diff --git a/src/backend/utils/adt/selfuncs.c b/src/backend/utils/adt/selfuncs.c
index d4da0e8dea9..f614a23d573 100644
--- a/src/backend/utils/adt/selfuncs.c
+++ b/src/backend/utils/adt/selfuncs.c
@@ -4394,7 +4394,8 @@ estimate_hash_bucket_stats(PlannerInfo *root, Node *hashkey, double nbuckets,
 {
 	VariableStatData vardata;
 	double		estfract,
-				ndistinct;
+				ndistinct,
+				stanullfrac;
 	bool		isdefault;
 	AttStatsSlot sslot;
 
@@ -4444,6 +4445,17 @@ estimate_hash_bucket_stats(PlannerInfo *root, Node *hashkey, double nbuckets,
 		return;
 	}
 
+	/* Get fraction that are null */
+	if (HeapTupleIsValid(vardata.statsTuple))
+	{
+		Form_pg_statistic stats;
+
+		stats = (Form_pg_statistic) GETSTRUCT(vardata.statsTuple);
+		stanullfrac = stats->stanullfrac;
+	}
+	else
+		stanullfrac = 0.0;
+
 	/*
 	 * Adjust ndistinct to account for restriction clauses.  Observe we are
 	 * assuming that the data distribution is affected uniformly by the
@@ -4468,6 +4480,13 @@ estimate_hash_bucket_stats(PlannerInfo *root, Node *hashkey, double nbuckets,
 	else
 		estfract = 1.0 / ndistinct;
 
+	/*
+	 * Adjust for null fraction.  NULL keys are not inserted into the hash
+	 * table, but inner_path_rows in final_cost_hashjoin includes them, so we
+	 * must discount estfract to compensate.
+	 */
+	estfract *= (1.0 - stanullfrac);
+
 	/*
 	 * Clamp the bucketsize fraction to be not less than the MCV frequency,
 	 * since whichever bucket the MCV values end up in will have at least that