diff --git a/src/odf/sbe/io.py b/src/odf/sbe/io.py
index 243d994..ce0247e 100644
--- a/src/odf/sbe/io.py
+++ b/src/odf/sbe/io.py
@@ -1,4 +1,3 @@
-from collections import Counter
 from hashlib import md5
 from pathlib import Path
 from typing import Literal
@@ -32,64 +31,54 @@ def string_writer(da: xr.DataArray, check=True) -> bytes:
     return _out
 
 
-def guess_scan_lengths(hex: str) -> int:
-    """Try to determine how many hex chars should be in each data line
-
-    If the number of bytes is in the header, return that * 2
-    If not, return the most common scan length
-    """
-    data = hex.lower()
-    d_split = data.splitlines()  # maybe this is expensive so only do it once
-    # data can be large, this header is probably in the first ~32k of data
-    if "number of bytes per scan" in data[:4096]:
-        for line in d_split:
-            if "number of bytes per scan" in line.lower():
-                return int(line.split("= ")[1]) * 2
-
-    counter = Counter(
-        len(line) for line in filter(lambda x: not x.startswith("*"), d_split)
-    )
-    return counter.most_common(1)[0][0]
-
-
 def hex_to_dataset(
     path: Path, errors: ERRORS = "store", encoding="CP437", content_md5=True
 ) -> xr.Dataset:
-    _comments = []  #   hex header comments written by deck box/SeaSave
-    out_idx = []  #   zero indexed "row" of the hex line, used for reconsturction of bad files
-    out = []  #   hex bytes out
+    out = bytearray()  #   hex bytes out
     hex = path.read_text(encoding)
 
-    error_idx = []
-    error_lines = []
-    linelen = guess_scan_lengths(hex) or 0
-    header_len = 0
-    for lineno, line in enumerate(hex.splitlines(), start=1):
-        if line.startswith("*"):  # comment
-            _comments.append(line)
-            header_len = lineno
-            continue
-
-        if len(line) != linelen:
-            if errors == "raise":
-                raise ValueError(f"invalid scan lengths line: {lineno}")
-            elif errors == "ignore":
-                continue
-            elif errors == "store":
-                error_idx.append(lineno - header_len)
-                error_lines.append(line)
-                continue
-
-        out_idx.append(lineno - header_len)
-        out.append([*bytes.fromhex(line)])
-    header = "\n".join(_comments)
-    data = np.array(out, dtype=np.uint8)
+    a = np.array(hex.splitlines())
+    comments_i = np.strings.startswith(a, "*")
+    data_i = ~comments_i
+
+    comments = a[comments_i]
+    data = a[data_i]
+    data_lens = np.strings.str_len(data)
+
+    # scan length?
+    scans = np.arange(data.shape[0]) + 1
+    comment_scan_length = (
+        np.strings.find(np.strings.lower(comments), "number of bytes per scan") > -1
+    )
+    # we want exactly one line that matches this
+    if comment_scan_length.sum() == 1:
+        line = comments[comment_scan_length]
+        line_length = int(line.item().split("= ")[1]) * 2
+    else:  # calcualte from the most common line length in data
+        counts_ = np.unique_counts(data_lens)
+        line_length = counts_.values[np.argmax(counts_.counts)]
+
+    ok_data_i = data_lens == line_length
+    ok_data = data[ok_data_i]
+    ok_scans = scans[ok_data_i]
+
+    bad_data_i = ~ok_data_i
+    bad_data = data[bad_data_i]
+    bad_scans = scans[bad_data_i]
+
+    if len(bad_data) > 0 and errors == "raise":
+        raise ValueError("invalid scan lengths")
+
+    out = np.frombuffer(bytes.fromhex("".join(ok_data)), dtype=np.uint8).reshape(
+        (ok_data.shape[0], line_length // 2)
+    )
 
     data_array = xr.DataArray(
-        data, dims=["scan", "bytes_per_scan"], coords={"scan": out_idx}
+        out, dims=["scan", "bytes_per_scan"], coords={"scan": ok_scans}
     )
     data_array.attrs["header"] = (
-        header  # utf8 needs to be encoded using .attrs["charset"] when written back out
+        "\n".join(comments)
+        # header  # utf8 needs to be encoded using .attrs["charset"] when written back out
     )
 
     data_array.attrs["filename"] = path.name
@@ -107,10 +96,10 @@ def hex_to_dataset(
     # This is about 3~4mb chunks uncompressed depending on how many channels there are
     data_ararys = {"hex": data_array}
 
-    if errors == "store" and len(error_lines) > 0:
+    if errors == "store" and len(bad_data) > 0:
         # make a string array of the bad lines
         error_data_array = xr.DataArray(
-            error_lines, dims=["scan_errors"], coords={"scan_errors": error_idx}
+            bad_data, dims=["scan_errors"], coords={"scan_errors": bad_scans}
         )
         error_data_array.encoding["zlib"] = True  # compress the data
         error_data_array.encoding["complevel"] = 6  # use compression level 6