diff --git a/tfjs-converter/python/tensorflowjs/converters/converter.py b/tfjs-converter/python/tensorflowjs/converters/converter.py
index c81167373e1..0bf06cd7928 100644
--- a/tfjs-converter/python/tensorflowjs/converters/converter.py
+++ b/tfjs-converter/python/tensorflowjs/converters/converter.py
@@ -390,7 +390,8 @@ def _standardize_input_output_formats(input_format, output_format):
   input_format_is_keras = (
       input_format in [common.KERAS_MODEL, common.KERAS_SAVED_MODEL])
   input_format_is_tf = (
-      input_format in [common.TF_SAVED_MODEL, common.TF_HUB_MODEL])
+      input_format in [common.TF_SAVED_MODEL,
+                       common.TF_FROZEN_MODEL, common.TF_HUB_MODEL])
   if output_format is None:
     # If no explicit output_format is provided, infer it from input format.
     if input_format_is_keras:
@@ -552,6 +553,10 @@ def convert(arguments):
       quantization.QUANTIZATION_BYTES_TO_DTYPES[args.quantization_bytes]
       if args.quantization_bytes else None)
 
+  if (not args.output_node_names and input_format == common.TF_FROZEN_MODEL):
+    raise ValueError(
+        'The --output_node_names flag is required for "tf_frozen_model"')
+
   if (args.signature_name and input_format not in
       (common.TF_SAVED_MODEL, common.TF_HUB_MODEL)):
     raise ValueError(