diff --git a/js/common/lib/tensor-impl-type-mapping.ts b/js/common/lib/tensor-impl-type-mapping.ts
index c4a43ea27fea1..b7cb176db393b 100644
--- a/js/common/lib/tensor-impl-type-mapping.ts
+++ b/js/common/lib/tensor-impl-type-mapping.ts
@@ -1,11 +1,17 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
+
 import {Tensor} from './tensor.js';
 
-export type SupportedTypedArrayConstructors = Float32ArrayConstructor|Uint8ArrayConstructor|Int8ArrayConstructor|
-    Uint16ArrayConstructor|Int16ArrayConstructor|Int32ArrayConstructor|BigInt64ArrayConstructor|Uint8ArrayConstructor|
-    Float64ArrayConstructor|Uint32ArrayConstructor|BigUint64ArrayConstructor;
+interface Float16ArrayConstructor {
+  new(): Float16Array;
+}
+
+export type SupportedTypedArrayConstructors = Float32ArrayConstructor|Float16ArrayConstructor|Uint8ArrayConstructor|
+    Int8ArrayConstructor|Uint16ArrayConstructor|Int16ArrayConstructor|Int32ArrayConstructor|BigInt64ArrayConstructor|
+    Uint8ArrayConstructor|Float64ArrayConstructor|Uint32ArrayConstructor|BigUint64ArrayConstructor;
 export type SupportedTypedArray = InstanceType<SupportedTypedArrayConstructors>;
 
 // a runtime map that maps type string to TypedArray constructor. Should match Tensor.DataTypeMap.
@@ -14,7 +20,7 @@ export const NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP = new Map<string, SupportedTy
   ['uint8', Uint8Array],
   ['int8', Int8Array],
   ['uint16', Uint16Array],
-  ['float16', Uint16Array],
+  ['float16', Float16Array],
   ['int16', Int16Array],
   ['int32', Int32Array],
   ['bool', Uint8Array],
diff --git a/js/common/lib/tensor-impl.ts b/js/common/lib/tensor-impl.ts
index e3e2b9c728556..9952567806987 100644
--- a/js/common/lib/tensor-impl.ts
+++ b/js/common/lib/tensor-impl.ts
@@ -1,6 +1,8 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {isFloat16Array} from '@petamoriken/float16';
+
 import {tensorToDataURL, tensorToImageData} from './tensor-conversion-impl.js';
 import {TensorToDataUrlOptions, TensorToImageDataOptions} from './tensor-conversion.js';
 import {tensorFromGpuBuffer, tensorFromImage, tensorFromPinnedBuffer, tensorFromTexture} from './tensor-factory-impl.js';
@@ -146,8 +148,8 @@ export class Tensor implements TensorInterface {
               // Throw error here because when user try to use number array as data,
               // e.g. new Tensor('float16', [1, 2, 3, 4], dims)), it will actually call
               // Uint16Array.from(arg1) which generates wrong data.
-              throw new TypeError(
-                  'Creating a float16 tensor from number array is not supported. Please use Uint16Array as data.');
+              // eslint-disable-next-line  @typescript-eslint/no-explicit-any
+              data = (typedArrayConstructor as any).from(arg1);
             } else if (arg0 === 'uint64' || arg0 === 'int64') {
               // use 'as any' here because:
               // 1. TypeScript's check on type of 'Array.isArray()' does not work with readonly arrays.
@@ -166,7 +168,7 @@ export class Tensor implements TensorInterface {
               // eslint-disable-next-line @typescript-eslint/no-explicit-any
               data = (typedArrayConstructor as any).from(arg1);
             }
-          } else if (arg1 instanceof typedArrayConstructor) {
+          } else if (arg1 instanceof typedArrayConstructor || isFloat16Array(arg1)) {
             data = arg1;
           } else {
             throw new TypeError(`A ${type} tensor's data must be type of ${typedArrayConstructor}`);
diff --git a/js/common/lib/tensor.ts b/js/common/lib/tensor.ts
index 6c08d1fe8e057..e6d4f3145a7e2 100644
--- a/js/common/lib/tensor.ts
+++ b/js/common/lib/tensor.ts
@@ -1,6 +1,8 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
+
 import {TensorFactory} from './tensor-factory.js';
 import {Tensor as TensorImpl} from './tensor-impl.js';
 import {TypedTensorUtils} from './tensor-utils.js';
@@ -74,7 +76,7 @@ export declare namespace Tensor {
     int64: BigInt64Array;
     string: string[];
     bool: Uint8Array;
-    float16: Uint16Array;  // Keep using Uint16Array until we have a concrete solution for float 16.
+    float16: Float16Array;  // Keep using Uint16Array until we have a concrete solution for float 16.
     float64: Float64Array;
     uint32: Uint32Array;
     uint64: BigUint64Array;
diff --git a/js/common/package.json b/js/common/package.json
index 64ab2736adbe3..0be6e5992daaf 100644
--- a/js/common/package.json
+++ b/js/common/package.json
@@ -30,5 +30,8 @@
     "ONNXRuntime",
     "ONNX Runtime"
   ],
-  "description": "ONNXRuntime JavaScript API library"
+  "description": "ONNXRuntime JavaScript API library",
+  "dependencies": {
+    "@petamoriken/float16": "^3.8.4"
+  }
 }
diff --git a/js/common/test/unit-tests/common.ts b/js/common/test/unit-tests/common.ts
index 49ebe872880a2..4fa86d43fbcdd 100644
--- a/js/common/test/unit-tests/common.ts
+++ b/js/common/test/unit-tests/common.ts
@@ -1,6 +1,7 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
 import assert from 'assert/strict';
 import {Tensor} from 'onnxruntime-common';
 
@@ -34,7 +35,7 @@ export const BIGINT_TYPES = [
 /**
  * float16 type, data represented by Uint16Array
  */
-export const FLOAT16_TYPE = ['float16', Uint16Array, false] as const;
+export const FLOAT16_TYPE = ['float16', Float16Array, false] as const;
 
 /**
  * A list of all numerical types.
diff --git a/js/web/lib/onnxjs/tensor.ts b/js/web/lib/onnxjs/tensor.ts
index 1a4c1dfe7494d..d5a4cf81f94cc 100644
--- a/js/web/lib/onnxjs/tensor.ts
+++ b/js/web/lib/onnxjs/tensor.ts
@@ -1,6 +1,7 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array, isFloat16Array} from '@petamoriken/float16';
 import {Guid} from 'guid-typescript';
 import Long from 'long';
 
@@ -13,6 +14,7 @@ import ortFbs = onnxruntime.experimental.fbs;
 export declare namespace Tensor {
   export interface DataTypeMap {
     bool: Uint8Array;
+    float16: Float16Array;
     float32: Float32Array;
     float64: Float64Array;
     string: string[];
@@ -31,7 +33,7 @@ export declare namespace Tensor {
   export type BooleanType = Tensor.DataTypeMap['bool'];
   export type IntegerType = Tensor.DataTypeMap['int8']|Tensor.DataTypeMap['uint8']|Tensor.DataTypeMap['int16']|
                             Tensor.DataTypeMap['uint16']|Tensor.DataTypeMap['int32']|Tensor.DataTypeMap['uint32'];
-  export type FloatType = Tensor.DataTypeMap['float32']|Tensor.DataTypeMap['float64'];
+  export type FloatType = Tensor.DataTypeMap['float16']|Tensor.DataTypeMap['float32']|Tensor.DataTypeMap['float64'];
   export type NumberType = BooleanType|IntegerType|FloatType;
 
   export type Id = Guid;
@@ -93,6 +95,7 @@ export class Tensor {
    */
   get floatData() {
     switch (this.type) {
+      case 'float16':
       case 'float32':
       case 'float64':
         return this.data as Tensor.FloatType;
@@ -188,7 +191,7 @@ export class Tensor {
     } else {
       if (cache !== undefined) {
         const constructor = dataviewConstructor(type);
-        if (!(cache instanceof constructor)) {
+        if (!(cache instanceof constructor) && !isFloat16Array(cache)) {
           throw new TypeError(`cache should be type ${constructor.name}`);
         }
       }
@@ -357,6 +360,7 @@ function sizeof(type: Tensor.DataType): number {
       return 1;
     case 'int16':
     case 'uint16':
+    case 'float16':
       return 2;
     case 'int32':
     case 'uint32':
@@ -412,6 +416,8 @@ function dataviewConstructor(type: Tensor.DataType) {
       return Uint32Array;
     case 'int64':
       return BigInt64Array;
+    case 'float16':
+      return Float16Array;
     case 'float32':
       return Float32Array;
     case 'float64':
diff --git a/js/web/lib/wasm/jsep/tensor-view.ts b/js/web/lib/wasm/jsep/tensor-view.ts
index 69b9287f6de29..f44f4efa21dbe 100644
--- a/js/web/lib/wasm/jsep/tensor-view.ts
+++ b/js/web/lib/wasm/jsep/tensor-view.ts
@@ -1,12 +1,13 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
 import {Tensor} from 'onnxruntime-common';
 
 import {tensorTypeToTypedArrayConstructor} from '../wasm-common';
 
 export const createView = (dataBuffer: ArrayBuffer, type: Tensor.Type): Int32Array|Uint32Array|BigInt64Array|
-    BigUint64Array|Uint8Array|Float32Array|Float64Array|Int8Array|Int16Array|Uint16Array =>
+    BigUint64Array|Uint8Array|Float16Array|Float32Array|Float64Array|Int8Array|Int16Array|Uint16Array =>
         new (tensorTypeToTypedArrayConstructor(type))(dataBuffer);
 
 /**
diff --git a/js/web/lib/wasm/wasm-common.ts b/js/web/lib/wasm/wasm-common.ts
index b9eff45e890c4..4227343390092 100644
--- a/js/web/lib/wasm/wasm-common.ts
+++ b/js/web/lib/wasm/wasm-common.ts
@@ -1,8 +1,13 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
 import {Tensor} from 'onnxruntime-common';
 
+interface Float16ArrayConstructor {
+  new(): Float16Array;
+}
+
 // This file includes common definitions. They do NOT have dependency on the WebAssembly instance.
 
 /**
@@ -112,12 +117,13 @@ export const getTensorElementSize = (dateType: number): number|
 /**
  * get typed array constructor by the given tensor type
  */
-export const tensorTypeToTypedArrayConstructor = (type: Tensor.Type): Float32ArrayConstructor|Uint8ArrayConstructor|
-    Int8ArrayConstructor|Uint16ArrayConstructor|Int16ArrayConstructor|Int32ArrayConstructor|BigInt64ArrayConstructor|
-    Uint8ArrayConstructor|Float64ArrayConstructor|Uint32ArrayConstructor|BigUint64ArrayConstructor => {
+export const tensorTypeToTypedArrayConstructor =
+    (type: Tensor.Type): Float16ArrayConstructor|Float32ArrayConstructor|Uint8ArrayConstructor|Int8ArrayConstructor|
+    Uint16ArrayConstructor|Int16ArrayConstructor|Int32ArrayConstructor|BigInt64ArrayConstructor|Uint8ArrayConstructor|
+    Float64ArrayConstructor|Uint32ArrayConstructor|BigUint64ArrayConstructor => {
       switch (type) {
         case 'float16':
-          return Uint16Array;
+          return Float16Array;
         case 'float32':
           return Float32Array;
         case 'uint8':
diff --git a/js/web/package.json b/js/web/package.json
index a502c2b6b032d..12e75faa137e9 100644
--- a/js/web/package.json
+++ b/js/web/package.json
@@ -11,6 +11,7 @@
   "version": "1.18.0",
   "jsdelivr": "dist/ort.min.js",
   "dependencies": {
+    "@petamoriken/float16": "^3.8.4",
     "flatbuffers": "^1.12.0",
     "guid-typescript": "^1.0.9",
     "long": "^5.2.3",
diff --git a/js/web/test/data/ops/padf16.jsonc b/js/web/test/data/ops/padf16.jsonc
new file mode 100644
index 0000000000000..e7b18281c5c66
--- /dev/null
+++ b/js/web/test/data/ops/padf16.jsonc
@@ -0,0 +1,35 @@
+[
+  {
+    "name": "constant 2D",
+    "operator": "Pad",
+    "opset": { "domain": "", "version": 10 },
+    "attributes": [
+      { "name": "mode", "data": "constant", "type": "string" },
+      { "name": "value", "data": 1.2, "type": "float" },
+      { "name": "pads", "data": [3, 2, 2, 3], "type": "ints" }
+    ],
+    "cases": [
+      {
+        "name": "[2,2]->[7,7]",
+        "inputs": [
+          {
+            "data": [1.0, 2.0, 3.0, 4.0],
+            "dims": [2, 2],
+            "type": "float16"
+          }
+        ],
+        "outputs": [
+          {
+            "data": [
+              1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2,
+              1.2, 1.2, 1.0, 2.0, 1.2, 1.2, 1.2, 1.2, 1.2, 3.0, 4.0, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2,
+              1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2
+            ],
+            "dims": [7, 7],
+            "type": "float16"
+          }
+        ]
+      }
+    ]
+  }
+]
diff --git a/js/web/test/test-runner.ts b/js/web/test/test-runner.ts
index 442cb1bcf1f34..2e804bc353a3e 100644
--- a/js/web/test/test-runner.ts
+++ b/js/web/test/test-runner.ts
@@ -1,6 +1,7 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
 import {expect} from 'chai';
 import * as ort from 'onnxruntime-common';
 import {extname} from 'path';
@@ -393,11 +394,12 @@ export class TensorResultValidator {
       case 'string':
         return this.strictEqual(actual.stringData, expected.stringData);
 
+      case 'float16':
       case 'float32':
       case 'float64':
         return this.floatEqual(
-            actual.numberData as number[] | Float32Array | Float64Array,
-            expected.numberData as number[] | Float32Array | Float64Array);
+            actual.numberData as number[] | Float16Array | Float32Array | Float64Array,
+            expected.numberData as number[] | Float16Array | Float32Array | Float64Array);
 
       case 'uint8':
       case 'int8':
@@ -425,7 +427,10 @@ export class TensorResultValidator {
       return false;
     }
   }
-  floatEqual(actual: number[]|Float32Array|Float64Array, expected: number[]|Float32Array|Float64Array): boolean {
+
+  floatEqual(
+      actual: number[]|Float16Array|Float32Array|Float64Array,
+      expected: number[]|Float16Array|Float32Array|Float64Array): boolean {
     if (actual.length !== expected.length) {
       return false;
     }
diff --git a/js/web/test/unittests/backends/webgl/test-conv-utils.ts b/js/web/test/unittests/backends/webgl/test-conv-utils.ts
index 32cace1ea9040..8e7f0f5d21b9a 100644
--- a/js/web/test/unittests/backends/webgl/test-conv-utils.ts
+++ b/js/web/test/unittests/backends/webgl/test-conv-utils.ts
@@ -1,14 +1,18 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
+import {Float16Array} from '@petamoriken/float16';
+
 import {Tensor} from '../../../../lib/onnxjs/tensor';
 
 /* eslint-disable no-bitwise */
 
+type FloatTypedArray = Float16Array|Float32Array|Float64Array;
+
 // eslint-disable-next-line no-underscore-dangle
 function matMul2d_(
-    A: Float32Array|Float64Array, B: Float32Array|Float64Array, C: Float32Array|Float64Array, alpha: number,
-    beta: number, M: number, N: number, K: number) {
+    A: FloatTypedArray, B: FloatTypedArray, C: FloatTypedArray, alpha: number, beta: number, M: number, N: number,
+    K: number) {
   let offsetA = 0, offsetB = 0, offsetC = 0;
   for (let mm = 0; mm < M; mm++) {
     for (let nn = 0; nn < N; nn++) {
@@ -30,8 +34,8 @@ function matMul2d_(
 }
 
 function matMul2d_tA(
-    A: Float32Array|Float64Array, B: Float32Array|Float64Array, C: Float32Array|Float64Array, alpha: number,
-    beta: number, M: number, N: number, K: number) {
+    A: FloatTypedArray, B: FloatTypedArray, C: FloatTypedArray, alpha: number, beta: number, M: number, N: number,
+    K: number) {
   let offsetA = 0, offsetB = 0, offsetC = 0;
   for (let mm = 0; mm < M; mm++) {
     for (let nn = 0; nn < N; nn++) {
@@ -53,8 +57,8 @@ function matMul2d_tA(
 }
 
 function matMul2d_tB(
-    A: Float32Array|Float64Array, B: Float32Array|Float64Array, C: Float32Array|Float64Array, alpha: number,
-    beta: number, M: number, N: number, K: number) {
+    A: FloatTypedArray, B: FloatTypedArray, C: FloatTypedArray, alpha: number, beta: number, M: number, N: number,
+    K: number) {
   let offsetA = 0, offsetB = 0, offsetC = 0;
   for (let mm = 0; mm < M; mm++) {
     for (let nn = 0; nn < N; nn++) {
@@ -76,8 +80,8 @@ function matMul2d_tB(
 }
 
 function matMul2d_tAtB(
-    A: Float32Array|Float64Array, B: Float32Array|Float64Array, C: Float32Array|Float64Array, alpha: number,
-    beta: number, M: number, N: number, K: number) {
+    A: FloatTypedArray, B: FloatTypedArray, C: FloatTypedArray, alpha: number, beta: number, M: number, N: number,
+    K: number) {
   let offsetA = 0, offsetB = 0, offsetC = 0;
   for (let mm = 0; mm < M; mm++) {
     for (let nn = 0; nn < N; nn++) {
@@ -105,8 +109,8 @@ function matMul2d_tAtB(
  * @param C data of tensor C, whose shape is [M,N]
  */
 export function matMul2d(
-    A: Float32Array|Float64Array, B: Float32Array|Float64Array, C: Float32Array|Float64Array, transA: boolean,
-    transB: boolean, alpha: number, beta: number, M: number, N: number, K: number): void {
+    A: FloatTypedArray, B: FloatTypedArray, C: FloatTypedArray, transA: boolean, transB: boolean, alpha: number,
+    beta: number, M: number, N: number, K: number): void {
   if (transA && transB) {
     matMul2d_tAtB(A, B, C, alpha, beta, M, N, K);
   } else if (transA) {
@@ -119,9 +123,9 @@ export function matMul2d(
 }
 
 function im2col(
-    data_im: Float32Array|Float64Array, data_col: Float32Array|Float64Array, channels: number, height: number,
-    width: number, kernel_h: number, kernel_w: number, dilation_h: number, dilation_w: number, pad_t: number,
-    pad_l: number, pad_b: number, pad_r: number, stride_h: number, stride_w: number) {
+    data_im: FloatTypedArray, data_col: FloatTypedArray, channels: number, height: number, width: number,
+    kernel_h: number, kernel_w: number, dilation_h: number, dilation_w: number, pad_t: number, pad_l: number,
+    pad_b: number, pad_r: number, stride_h: number, stride_w: number) {
   const output_h = ~~((height + pad_b + pad_t - (dilation_h * (kernel_h - 1) + 1)) / stride_h) + 1;
   const output_w = ~~((width + pad_l + pad_r - (dilation_w * (kernel_w - 1) + 1)) / stride_w) + 1;