diff --git a/doc/source/whatsnew/v0.21.0.txt b/doc/source/whatsnew/v0.21.0.txt
index 5146bd35dff30..1c4effc31833f 100644
--- a/doc/source/whatsnew/v0.21.0.txt
+++ b/doc/source/whatsnew/v0.21.0.txt
@@ -201,7 +201,8 @@ I/O
 ^^^
 
 - Bug in :func:`read_csv` in which non integer values for the header argument generated an unhelpful / unrelated error message (:issue:`16338`)
-
+- Bug in :func:`read_csv` in which passing a CSV with at least one very large (i.e. more than 2^31 - 1 bytes) column along with ``low_memory=False`` would cause an integer overflow. The result was an always unsuccessful attempt to allocate an enourmous buffer and then reporting "Out of memory." (:issue:`16798`).
+- Bug in :func:`read_csv` in which some errors paths were assigning error messages to the internal tokenizer's ``error_msg`` field without first allocating the memory. When this happened as part of exception handling, it resulted in a double ``free`` and the program halted due to a ``SIGSEGV`` (:issue:`16798`).
 - Bug in :func:`read_stata` where value labels could not be read when using an iterator (:issue:`16923`)
 
 Plotting
diff --git a/pandas/_libs/parsers.pyx b/pandas/_libs/parsers.pyx
index 88c695a3faf27..72e5fb7f12c7f 100644
--- a/pandas/_libs/parsers.pyx
+++ b/pandas/_libs/parsers.pyx
@@ -137,7 +137,7 @@ cdef extern from "parser/tokenizer.h":
         int64_t words_len
         int64_t words_cap
 
-        char *pword_start       # pointer to stream start of current field
+        char *pword_start        # pointer to stream start of current field
         int64_t word_start       # position start of current field
 
         int64_t *line_start      # position in words for start of line
@@ -177,9 +177,9 @@ cdef extern from "parser/tokenizer.h":
         # thousands separator (comma, period)
         char thousands
 
-        int header # Boolean: 1: has header, 0: no header
-        int64_t header_start # header row start
-        int64_t header_end # header row end
+        int header                  # Boolean: 1: has header, 0: no header
+        int64_t header_start        # header row start
+        int64_t header_end          # header row end
 
         void *skipset
         PyObject *skipfunc