diff --git a/rockfish/rftools/merge.py b/rockfish/rftools/merge.py
new file mode 100644
index 0000000..bfde7ee
--- /dev/null
+++ b/rockfish/rftools/merge.py
@@ -0,0 +1,113 @@
+from tqdm import tqdm
+
+from dataclasses import dataclass
+import sys
+from io import BufferedReader, BufferedWriter
+import struct
+from collections import defaultdict
+from itertools import count
+import argparse
+
+from typing import *
+
+
+@dataclass
+class RFHeader:
+    ctgs: List[int]
+    n_examples: int
+    start_offset: int
+
+
+def parse_ctgs(fd: BufferedReader) -> List[str]:
+    fd.seek(0)
+
+    n_ctgs = int.from_bytes(fd.read(2), byteorder=sys.byteorder)
+    ctgs = []
+    for _ in range(n_ctgs):
+        ctg_name_len = int.from_bytes(fd.read(1), byteorder=sys.byteorder)
+        ctg_name = struct.unpack(f'={ctg_name_len}s', fd.read(ctg_name_len))[0]
+        ctgs.append(ctg_name.decode())
+
+    return ctgs
+
+
+def map_header(src: List[str]) -> Tuple[Dict[str, int], Dict[str, RFHeader]]:
+    ctg_encoding = defaultdict(lambda c=count(): next(c))
+    header_info = []
+
+    for rf_file in src:
+        with open(rf_file, 'rb') as rf_src:
+            c_map = [ctg_encoding[c] for c in parse_ctgs(rf_src)]
+            n_examples = int.from_bytes(rf_src.read(4),
+                                        byteorder=sys.byteorder,
+                                        signed=False)
+            start_offset = rf_src.tell()
+
+        header_info.append(RFHeader(c_map, n_examples, start_offset))
+
+    return ctg_encoding, header_info
+
+
+def write_header(dest: BufferedWriter, ctg_encoding: Dict[str, int],
+                 n_examples: int) -> int:
+    # Writing ctgs + placeholder for number of examples
+    n_refs = len(ctg_encoding)
+    data = struct.pack('=H', n_refs)
+
+    for ref_name, _ in ctg_encoding.items():
+        ref_len = len(ref_name)
+        data += struct.pack(f'=B{ref_len}s', ref_len, str.encode(ref_name))
+
+    n_examples_offset = len(data)
+    data += struct.pack('=I', n_examples)
+    dest.write(data)
+
+    return n_examples_offset
+
+
+def merge(src: List[str], dest: str, seq_len: int) -> None:
+    tqdm.write('Reading source headers.')
+
+    ctg_encoding, header_info = map_header(src)
+    n_examples = sum([h.n_examples for h in header_info])
+    tqdm.write(f'Total {n_examples} examples found.')
+
+    with open(dest, 'wb') as rf_dest:
+        write_header(rf_dest, ctg_encoding, n_examples)
+        tqdm.write('Destination header written.\nWriting examples...')
+
+        for rf_file, header in zip(tqdm(src), header_info):
+            with open(rf_file, 'rb') as rf_src:
+                rf_src.seek(header.start_offset)
+
+                for _ in range(header.n_examples):
+                    example_info = rf_src.read(46)
+                    _, _, _, n_points, q_indices_len = struct.unpack(
+                        '=36sHIHH', example_info)
+                    n_bytes = 2 * n_points + 2 * q_indices_len + 3 * seq_len
+                    example_data = rf_src.read(n_bytes)
+
+                    rf_dest.write(example_info + example_data)
+
+                # assert EOF
+                if rf_src.read():
+                    tqdm.write(
+                        f'ERROR: {rf_file} EOF expected but data was read.')
+                    sys.exit(1)
+
+    tqdm.write('Processing finished.')
+
+
+def parse_arguments() -> argparse.Namespace:
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument('src', type=str, nargs='+')
+    parser.add_argument('dest', type=str)
+    parser.add_argument('-l', '--seq_len', type=int, default=31)
+
+    return parser.parse_args()
+
+
+if __name__ == '__main__':
+    args = parse_arguments()
+    merge(args.src, args.dest, args.seq_len)