diff --git a/.circleci/unittest/linux/scripts/environment.yml b/.circleci/unittest/linux/scripts/environment.yml
index b2efb93de75..8bec8b2b054 100644
--- a/.circleci/unittest/linux/scripts/environment.yml
+++ b/.circleci/unittest/linux/scripts/environment.yml
@@ -6,8 +6,7 @@ dependencies:
   - pytest-cov
   - pip
   - libpng
-  # NOTE: Pinned to fix issues with size_t on Windows
-  - jpeg <=9b
+  - jpeg
   - ca-certificates
   - pip:
     - future
diff --git a/.circleci/unittest/windows/scripts/environment.yml b/.circleci/unittest/windows/scripts/environment.yml
index da2f502f3fe..02ac1285dea 100644
--- a/.circleci/unittest/windows/scripts/environment.yml
+++ b/.circleci/unittest/windows/scripts/environment.yml
@@ -6,8 +6,7 @@ dependencies:
   - pytest-cov
   - pip
   - libpng
-  # NOTE: Pinned to fix issues with size_t on Windows
-  - jpeg <=9b
+  - jpeg
   - ca-certificates
   - pip:
     - future
diff --git a/packaging/torchvision/meta.yaml b/packaging/torchvision/meta.yaml
index 222071ede93..da061c5cc2a 100644
--- a/packaging/torchvision/meta.yaml
+++ b/packaging/torchvision/meta.yaml
@@ -9,8 +9,7 @@ requirements:
   build:
     - {{ compiler('c') }} # [win]
     - libpng
-    # NOTE: Pinned to fix issues with size_t on Windows
-    - jpeg <=9b
+    - jpeg
     # NOTE: The only ffmpeg version that we build is actually 4.2
     - ffmpeg >=4.2  # [not win]
 
@@ -25,8 +24,7 @@ requirements:
     - python
     - libpng
     - ffmpeg >=4.2  # [not win]
-    # NOTE: Pinned to fix issues with size_t on Windows
-    - jpeg <=9b
+    - jpeg
     - pillow >=5.3.0
     {{ environ.get('CONDA_PYTORCH_CONSTRAINT') }}
     {{ environ.get('CONDA_CUDATOOLKIT_CONSTRAINT') }}
@@ -52,8 +50,7 @@ test:
   requires:
     - pytest
     - scipy
-    # NOTE: Pinned to fix issues with size_t on Windows
-    - jpeg <=9b
+    - jpeg
     - ca-certificates
 
 
diff --git a/torchvision/csrc/io/image/cpu/encode_jpeg.cpp b/torchvision/csrc/io/image/cpu/encode_jpeg.cpp
index c84ad37005d..2f445d21e47 100644
--- a/torchvision/csrc/io/image/cpu/encode_jpeg.cpp
+++ b/torchvision/csrc/io/image/cpu/encode_jpeg.cpp
@@ -13,6 +13,18 @@ torch::Tensor encode_jpeg(const torch::Tensor& data, int64_t quality) {
 }
 
 #else
+// For libjpeg version <= 9b, the out_size parameter in jpeg_mem_dest() is
+// defined as unsigned long, where as in later version, it is defined as size_t.
+// For windows backward compatibility, we define JpegSizeType as different types
+// according to the libjpeg version used, in order to prevent compilcation
+// errors.
+#if defined(_WIN32) || !defined(JPEG_LIB_VERSION_MAJOR) || \
+    (JPEG_LIB_VERSION_MAJOR < 9) ||                        \
+    (JPEG_LIB_VERSION_MAJOR == 9 && JPEG_LIB_VERSION_MINOR <= 2)
+using JpegSizeType = unsigned long;
+#else
+using JpegSizeType = size_t;
+#endif
 
 using namespace detail;
 
@@ -22,7 +34,7 @@ torch::Tensor encode_jpeg(const torch::Tensor& data, int64_t quality) {
   struct torch_jpeg_error_mgr jerr;
 
   // Define buffer to write JPEG information to and its size
-  unsigned long jpegSize = 0;
+  JpegSizeType jpegSize = 0;
   uint8_t* jpegBuf = NULL;
 
   cinfo.err = jpeg_std_error(&jerr.pub);