diff --git a/stac-arrow/Cargo.toml b/stac-arrow/Cargo.toml
index c96280493..fa7f9fd6c 100644
--- a/stac-arrow/Cargo.toml
+++ b/stac-arrow/Cargo.toml
@@ -19,5 +19,10 @@ thiserror = "1"
 wkb = "0.7"
 
 [dev-dependencies]
+criterion = { version = "0.5", features = ["html_reports"] }
 parquet = "51"
 stac-validate = { version = "0.1", path = "../stac-validate" }
+
+[[bench]]
+name = "read"
+harness = false
diff --git a/stac-arrow/benches/read.rs b/stac-arrow/benches/read.rs
new file mode 100644
index 000000000..66f7fab66
--- /dev/null
+++ b/stac-arrow/benches/read.rs
@@ -0,0 +1,41 @@
+//! How fast is `RecordBatch` -> `Vec<Map<String, Value>>` when going through
+//! full serialization vs the deprecated `record_batches_to_json_rows`?
+
+use arrow::array::RecordBatch;
+use arrow_json::ArrayWriter;
+use criterion::{criterion_group, criterion_main, Criterion};
+use parquet::arrow::arrow_reader::ParquetRecordBatchReaderBuilder;
+use serde_json::{Map, Value};
+use std::fs::File;
+
+fn record_batches_to_json_rows(record_batch: &RecordBatch) {
+    let mut writer = ArrayWriter::new(Vec::new());
+    writer.write(record_batch).unwrap();
+    writer.finish().unwrap();
+    let _: Vec<Map<String, Value>> =
+        serde_json::from_reader(writer.into_inner().as_slice()).unwrap();
+}
+
+#[allow(deprecated)]
+fn writer(record_batch: &RecordBatch) {
+    let _ = arrow_json::writer::record_batches_to_json_rows(&[record_batch]).unwrap();
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let mut group = c.benchmark_group("record batch to vec");
+    let file = File::open("data/naip.parquet").unwrap();
+    let mut reader = ParquetRecordBatchReaderBuilder::try_new(file)
+        .unwrap()
+        .build()
+        .unwrap();
+    let mut record_batch = reader.next().unwrap().unwrap();
+    let index = record_batch.schema().index_of("geometry").unwrap();
+    record_batch.remove_column(index);
+    group.bench_function("record_batches_to_json_rows", |b| {
+        b.iter(|| record_batches_to_json_rows(&record_batch))
+    });
+    group.bench_function("writer", |b| b.iter(|| writer(&record_batch)));
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);
diff --git a/stac-arrow/src/lib.rs b/stac-arrow/src/lib.rs
index 23fe54a88..dbc55ae07 100644
--- a/stac-arrow/src/lib.rs
+++ b/stac-arrow/src/lib.rs
@@ -221,3 +221,6 @@ mod readme {
 
     external_doc_test!(include_str!("../README.md"));
 }
+
+#[cfg(test)]
+use criterion as _;