diff --git a/mutation_query.cc b/mutation_query.cc
index 3ab0d2145a..ff6379d394 100644
--- a/mutation_query.cc
+++ b/mutation_query.cc
@@ -57,13 +57,14 @@ bool reconcilable_result::operator!=(const reconcilable_result& other) const {
 }
 
 query::result
-to_data_query_result(const reconcilable_result& r, schema_ptr s, const query::partition_slice& slice, uint32_t max_partitions) {
+to_data_query_result(const reconcilable_result& r, schema_ptr s, const query::partition_slice& slice, uint32_t max_rows, uint32_t max_partitions) {
     query::result::builder builder(slice, query::result_request::only_result, { });
     for (const partition& p : r.partitions()) {
-        if (!max_partitions--) {
+        if (builder.row_count() >= max_rows || builder.partition_count() >= max_partitions) {
             break;
         }
-        p.mut().unfreeze(s).query(builder, slice, gc_clock::time_point::min(), query::max_rows);
+        // Also enforces the per-partition limit.
+        p.mut().unfreeze(s).query(builder, slice, gc_clock::time_point::min(), max_rows - builder.row_count());
     }
     if (r.is_short_read()) {
         builder.mark_as_short_read();
diff --git a/mutation_query.hh b/mutation_query.hh
index 35db45a53f..2009747d88 100644
--- a/mutation_query.hh
+++ b/mutation_query.hh
@@ -105,7 +105,7 @@ public:
     printer pretty_printer(schema_ptr) const;
 };
 
-query::result to_data_query_result(const reconcilable_result&, schema_ptr, const query::partition_slice&, uint32_t partition_limit = query::max_partitions);
+query::result to_data_query_result(const reconcilable_result&, schema_ptr, const query::partition_slice&, uint32_t row_limit, uint32_t partition_limit);
 
 // Performs a query on given data source returning data in reconcilable form.
 //
diff --git a/service/storage_proxy.cc b/service/storage_proxy.cc
index 6c764c0a09..6c1d07c69e 100644
--- a/service/storage_proxy.cc
+++ b/service/storage_proxy.cc
@@ -2346,7 +2346,8 @@ protected:
                 if (rr_opt && (can_send_short_read || data_resolver->all_reached_end() || rr_opt->row_count() >= original_row_limit()
                                || data_resolver->live_partition_count() >= original_partition_limit())
                         && !data_resolver->any_partition_short_read()) {
-                    auto result = ::make_foreign(::make_lw_shared(to_data_query_result(std::move(*rr_opt), _schema, _cmd->slice)));
+                    auto result = ::make_foreign(::make_lw_shared(
+                            to_data_query_result(std::move(*rr_opt), _schema, _cmd->slice, _cmd->row_limit, cmd->partition_limit)));
                     // wait for write to complete before returning result to prevent multiple concurrent read requests to
                     // trigger repair multiple times and to prevent quorum read to return an old value, even after a quorum
                     // another read had returned a newer value (but the newer value had not yet been sent to the other replicas)