diff --git a/0x-segmented-file.md b/0x-segmented-file.md
new file mode 100644
index 0000000..79e7eae
--- /dev/null
+++ b/0x-segmented-file.md
@@ -0,0 +1,35 @@
+# Segmented file protocol
+
+The idea of the segmented file protocol is to have a standard way to put files on the blockchain in OP_RETURN data, where the file is separated over multiple transactions or outputs.
+
+## Protocol
+
+A high level outline of a segmented file is as follows:
+
+"sgfl" [filename] [data-reference]
+
+In detail these are:
+
+* "sgfl": A 4-byte prefix. In hex, it is 0x7367666c.
+* [filename]: Pushdata containing a byte string. However, it is recommended that the filename be a utf8 character string. Furthermore, it is recommended that it be divided into name and extension like normal files: [name].[extension] . Operating systems are already used to dealing with filenames and parsing meaning from them, so we can reuse code from operating systems for this.
+* [data-reference]: Pushdata containing a concatenated stream of, for each segment, a txid (32 bytes) and a varint tx output (variable length).
+
+An example of a segmented file is as follows:
+
+```
+OP_RETURN 4 0x7367666c 9 0x626c6164652e6a7067 103 22cd5757d5194daa014944128cbf86261a5f42e78c2f4e541b6b7774297e74b90222cd5757d5194daa014944128cbf86261a5f42e78c2f4e541b6b7774297e74b9fd122722cd5757d5194daa014944128cbf86261a5f42e78c2f4e541b6b7774297e74b9fd0002
+```
+
+This is the file 'blade.jpg' comprised of three parts, all in the same transaction, but in segments located in outputs 2, 10002 and 512.
+
+## Segment contents
+
+These are the ways segments are processed:
+
+* If the segment is recognised as being defined by this segmented file protocol, the reconstructed data is used as the segment data.
+* If the segment is recognised as being defined by the file protocol, the embedded file data is used as the segment data.
+* Otherwise the data within the first pushdata in the transaction output references as the segment, is used as the segment data.
+
+## On disk representation
+
+The complete OP_RETURN contents (excluding the leading OP_RETURN opcode) is stored in an on-disk file, perhaps as '[filename].bitcoin'.  It can be identified and processed using the same processing that identifies and processes the contents of an OP_RETURN.
\ No newline at end of file