diff --git a/RELEASE.md b/RELEASE.md
index c1b5095..f89db2a 100644
--- a/RELEASE.md
+++ b/RELEASE.md
@@ -6,6 +6,10 @@
 
 ## Bug Fixes and Other Changes
 
+*   For nested features with N nested levels (N > 1), the statistics counting
+    the number of values in `CommonStatistics` and `WeightedCommonStatistics`
+    will rely on the innermost level.
+
 ## Breaking Changes
 
 ## Deprecations
@@ -31,6 +35,8 @@
 
 *   Removed `NaturalLanguageDomain.location_constraint_regex`.
     It was documented as "please do not use" and never implemented.
+*   Change to the semantics of min/max/avg/tot num-values for nested features
+    (see above).
 
 ## Deprecations
 
diff --git a/tensorflow_metadata/proto/v0/statistics.proto b/tensorflow_metadata/proto/v0/statistics.proto
index 5f27247..79c785f 100644
--- a/tensorflow_metadata/proto/v0/statistics.proto
+++ b/tensorflow_metadata/proto/v0/statistics.proto
@@ -201,7 +201,8 @@ message FeatureNameStatistics {
 // Common weighted statistics for all feature types. Statistics counting number
 // of values (i.e., avg_num_values and tot_num_values) include NaNs.
 // If the weighted column is missing, then this counts as a weight of 1
-// for that example.
+// for that example. For nested features with N nested levels (N > 1), the
+// statistics counting number of values will rely on the innermost level.
 message WeightedCommonStatistics {
   // Weighted number of examples not missing.
   double num_non_missing = 1;
@@ -210,9 +211,9 @@ message WeightedCommonStatistics {
   // as missing.
   double num_missing = 2;
   // average number of values, weighted by the number of examples.
+  // avg_num_values = tot_num_values / num_non_missing.
   double avg_num_values = 3;
-  // tot_num_values = avg_num_values * num_non_missing.
-  // This is calculated directly, so should have less numerical error.
+  // The total number of values in this feature.
   double tot_num_values = 4;
 }
 
@@ -440,7 +441,8 @@ message PresenceAndValencyStatistics {
 
 // Common statistics for all feature types. Statistics counting number of values
 // (i.e., min_num_values, max_num_values, avg_num_values, and tot_num_values)
-// include NaNs.
+// include NaNs. For nested features with N nested levels (N > 1), the
+// statistics counting number of values will rely on the innermost level.
 message CommonStatistics {
   // The number of examples that include this feature. Note that this includes
   // examples that contain this feature with an explicitly empty list of values,
@@ -453,9 +455,9 @@ message CommonStatistics {
   // The maximum number of values in a single example for this feature.
   uint64 max_num_values = 4;
   // The average number of values in a single example for this feature.
+  // avg_num_values = tot_num_values / num_non_missing.
   float avg_num_values = 5;
-  // tot_num_values = avg_num_values * num_non_missing.
-  // This is calculated directly, so should have less numerical error.
+  // The total number of values in this feature.
   uint64 tot_num_values = 8;
   // The quantiles histogram for the number of values in this feature.
   Histogram num_values_histogram = 6;