diff --git a/be/src/vec/exec/format/parquet/parquet_pred_cmp.h b/be/src/vec/exec/format/parquet/parquet_pred_cmp.h
index 60001f649e..8d3057312b 100644
--- a/be/src/vec/exec/format/parquet/parquet_pred_cmp.h
+++ b/be/src/vec/exec/format/parquet/parquet_pred_cmp.h
@@ -369,13 +369,22 @@ private:
     }
 
 public:
-    static bool filter_by_min_max(const ColumnValueRangeType& col_val_range,
-                                  const FieldSchema* col_schema, const std::string& encoded_min,
-                                  const std::string& encoded_max, const cctz::time_zone& ctz) {
+    static bool filter_by_stats(const ColumnValueRangeType& col_val_range,
+                                const FieldSchema* col_schema, bool is_set_min_max,
+                                const std::string& encoded_min, const std::string& encoded_max,
+                                bool is_all_null, const cctz::time_zone& ctz) {
         bool need_filter = false;
         std::visit(
                 [&](auto&& range) {
                     std::vector<ScanPredicate> filters = _value_range_to_predicate(range);
+                    // Currently, ScanPredicate doesn't include "is null" && "x = null", filters will be empty when contains these exprs.
+                    // So we can handle is_all_null safely.
+                    if (!filters.empty()) {
+                        need_filter = is_all_null;
+                        if (need_filter) {
+                            return;
+                        }
+                    }
                     for (auto& filter : filters) {
                         need_filter |= _filter_by_min_max(range, filter, col_schema, encoded_min,
                                                           encoded_max, ctz);
diff --git a/be/src/vec/exec/format/parquet/vparquet_page_index.cpp b/be/src/vec/exec/format/parquet/vparquet_page_index.cpp
index bad05728eb..35cf076318 100644
--- a/be/src/vec/exec/format/parquet/vparquet_page_index.cpp
+++ b/be/src/vec/exec/format/parquet/vparquet_page_index.cpp
@@ -67,9 +67,10 @@ Status PageIndex::collect_skipped_page_range(tparquet::ColumnIndex* column_index
 
     const int num_of_pages = column_index->null_pages.size();
     for (int page_id = 0; page_id < num_of_pages; page_id++) {
-        if (ParquetPredicate::filter_by_min_max(col_val_range, col_schema,
-                                                encoded_min_vals[page_id],
-                                                encoded_max_vals[page_id], ctz)) {
+        bool is_all_null = column_index->null_pages[page_id];
+        if (ParquetPredicate::filter_by_stats(col_val_range, col_schema, !is_all_null,
+                                              encoded_min_vals[page_id], encoded_max_vals[page_id],
+                                              is_all_null, ctz)) {
             skipped_ranges.emplace_back(page_id);
         }
     }
diff --git a/be/src/vec/exec/format/parquet/vparquet_reader.cpp b/be/src/vec/exec/format/parquet/vparquet_reader.cpp
index efe81a953f..388dbcf838 100644
--- a/be/src/vec/exec/format/parquet/vparquet_reader.cpp
+++ b/be/src/vec/exec/format/parquet/vparquet_reader.cpp
@@ -801,14 +801,19 @@ Status ParquetReader::_process_column_stat_filter(const std::vector<tparquet::Co
             continue;
         }
         int parquet_col_id = col_iter->second;
-        auto& statistic = columns[parquet_col_id].meta_data.statistics;
-        if (!statistic.__isset.max || !statistic.__isset.min) {
+        auto& meta_data = columns[parquet_col_id].meta_data;
+        auto& statistic = meta_data.statistics;
+        bool is_all_null =
+                (statistic.__isset.null_count && statistic.null_count == meta_data.num_values);
+        bool is_set_min_max = (statistic.__isset.max && statistic.__isset.min);
+        if ((!is_set_min_max) && (!is_all_null)) {
             continue;
         }
         const FieldSchema* col_schema = schema_desc.get_column(col_name);
         // Min-max of statistic is plain-encoded value
-        *filter_group = ParquetPredicate::filter_by_min_max(slot_iter->second, col_schema,
-                                                            statistic.min, statistic.max, *_ctz);
+        *filter_group =
+                ParquetPredicate::filter_by_stats(slot_iter->second, col_schema, is_set_min_max,
+                                                  statistic.min, statistic.max, is_all_null, *_ctz);
         if (*filter_group) {
             break;
         }