diff --git a/crates/paimon/src/arrow/format/mosaic.rs b/crates/paimon/src/arrow/format/mosaic.rs
index 963226b8..23f97803 100644
--- a/crates/paimon/src/arrow/format/mosaic.rs
+++ b/crates/paimon/src/arrow/format/mosaic.rs
@@ -22,7 +22,8 @@ use crate::io::FileRead;
 use crate::spec::{DataField, DataType as PaimonDataType, Datum, Predicate};
 use crate::table::{ArrowRecordBatchStream, RowRange};
 use crate::Error;
-use arrow_array::{ArrayRef, RecordBatch, RecordBatchOptions, UInt64Array};
+use arrow_array::RecordBatch;
+use arrow_array::RecordBatchOptions;
 use arrow_schema::{DataType as ArrowDataType, SchemaRef, TimeUnit};
 use async_stream::try_stream;
 use async_trait::async_trait;
@@ -90,7 +91,7 @@ impl FormatFileReader for MosaicFormatReader {
                 let row_group_rows = mosaic_reader
                     .row_group_num_rows(row_group_index)
                     .map_err(mosaic_read_error)?;
-                let selected_indices = selected_indices_for_row_group(
+                let selected_slices = selected_slices_for_row_group(
                     row_group_rows,
                     row_group_start,
                     row_selection.as_deref(),
@@ -102,8 +103,8 @@ impl FormatFileReader for MosaicFormatReader {
                         source: None,
                     })?;
 
-                if let Some(indices) = selected_indices.as_ref() {
-                    if indices.is_empty() {
+                if let Some(slices) = selected_slices.as_ref() {
+                    if slices.is_empty() {
                         continue;
                     }
                 }
@@ -124,9 +125,9 @@ impl FormatFileReader for MosaicFormatReader {
                 }
 
                 let batch = if all_projected_columns_missing {
-                    let row_count = selected_indices
+                    let row_count = selected_slices
                         .as_ref()
-                        .map_or(row_group_rows, UInt64Array::len);
+                        .map_or(row_group_rows, |slices| selected_row_count(slices));
                     empty_batch(read_schema.clone(), row_count)?
                 } else {
                     let names = projected_names
@@ -140,7 +141,7 @@ impl FormatFileReader for MosaicFormatReader {
                     let batch = row_group_reader
                         .read_columns()
                         .map_err(mosaic_read_error)?;
-                    take_rows(batch, selected_indices.as_ref(), &read_schema)?
+                    take_row_slices(batch, selected_slices.as_deref(), &read_schema)?
                 };
                 for chunk in split_batch(batch, batch_size) {
                     yield chunk;
@@ -397,11 +398,11 @@ fn is_timestamp_nanos_struct(fields: &arrow_schema::Fields) -> bool {
         && *fields[1].data_type() == ArrowDataType::Int32
 }
 
-fn selected_indices_for_row_group(
+fn selected_slices_for_row_group(
     row_group_rows: usize,
     row_group_start: usize,
     row_selection: Option<&[RowRange]>,
-) -> crate::Result<Option<UInt64Array>> {
+) -> crate::Result<Option<Vec<(usize, usize)>>> {
     let Some(row_selection) = row_selection else {
         return Ok(None);
     };
@@ -414,7 +415,7 @@ fn selected_indices_for_row_group(
                 source: None,
             })?;
 
-    let mut indices = Vec::new();
+    let mut slices = Vec::new();
     for range in row_selection {
         let from = usize::try_from(range.from()).map_err(|e| Error::DataInvalid {
             message: format!(
@@ -438,41 +439,43 @@ fn selected_indices_for_row_group(
         if start >= end {
             continue;
         }
-        indices.extend((start - row_group_start..end - row_group_start).map(|idx| idx as u64));
+        slices.push((start - row_group_start, end - start));
     }
 
-    Ok(Some(UInt64Array::from(indices)))
+    Ok(Some(slices))
 }
 
-fn take_rows(
+fn selected_row_count(slices: &[(usize, usize)]) -> usize {
+    slices.iter().map(|(_, len)| *len).sum()
+}
+
+fn take_row_slices(
     batch: RecordBatch,
-    indices: Option<&UInt64Array>,
+    slices: Option<&[(usize, usize)]>,
     target_schema: &SchemaRef,
 ) -> crate::Result<RecordBatch> {
-    let Some(indices) = indices else {
+    let Some(slices) = slices else {
         return ensure_schema(batch, target_schema);
     };
 
-    if batch.num_columns() == 0 {
-        return empty_batch(target_schema.clone(), indices.len());
+    if slices.is_empty() {
+        return empty_batch(target_schema.clone(), 0);
     }
 
-    let columns = batch
-        .columns()
-        .iter()
-        .map(|column| {
-            arrow_select::take::take(column.as_ref(), indices, None).map_err(|e| {
-                Error::UnexpectedError {
-                    message: format!("Failed to apply Mosaic row selection: {e}"),
-                    source: Some(Box::new(e)),
-                }
-            })
-        })
-        .collect::<crate::Result<Vec<ArrayRef>>>()?;
+    if slices.len() == 1 {
+        let (offset, len) = slices[0];
+        return ensure_schema(batch.slice(offset, len), target_schema);
+    }
 
-    RecordBatch::try_new(target_schema.clone(), columns).map_err(|e| Error::UnexpectedError {
-        message: format!("Failed to build Mosaic RecordBatch: {e}"),
-        source: Some(Box::new(e)),
+    let sliced_batches = slices
+        .iter()
+        .map(|(offset, len)| batch.slice(*offset, *len))
+        .collect::<Vec<_>>();
+    arrow_select::concat::concat_batches(target_schema, &sliced_batches).map_err(|e| {
+        Error::UnexpectedError {
+            message: format!("Failed to apply Mosaic row selection: {e}"),
+            source: Some(Box::new(e)),
+        }
     })
 }
 
@@ -854,6 +857,24 @@ mod tests {
         assert_eq!(ids.values(), &[2, 3, 5]);
     }
 
+    #[test]
+    fn test_large_row_selection_stays_as_ranges() {
+        let row_group_rows = 1_000_000;
+        let selected = selected_slices_for_row_group(
+            row_group_rows,
+            0,
+            Some(&[
+                RowRange::new(0, 499_999),
+                RowRange::new(500_001, row_group_rows as i64 - 1),
+            ]),
+        )
+        .unwrap()
+        .unwrap();
+
+        assert_eq!(selected, vec![(0, 500_000), (500_001, 499_999)]);
+        assert_eq!(selected_row_count(&selected), row_group_rows - 1);
+    }
+
     #[tokio::test]
     async fn test_read_predicate_prunes_non_matching_row_groups() {
         let fields = data_fields();