diff --git a/README.md b/README.md
index 2f70fb7..53f942a 100644
--- a/README.md
+++ b/README.md
@@ -47,7 +47,7 @@ If you want to install it in other location:
 asciinema cli is composed of multiple (sub)commands, similar to `git`,
 `rails` or `brew`.
 
-If you run `asciinema` with no arguments, help will be displayed showing all
+If you run `asciinema` with no arguments help message is displayed showing all
 available commands.
 
 In addition to this, you can run any asciinema command with the `-h` switch
@@ -71,11 +71,11 @@ Recording finishes when you exit the shell (hit <kbd>Ctrl+D</kbd> or type
 the process exits.
 
 If the `filename` argument is given then the resulting recording (called
-*asciicast*) is saved to a local file. It can later be replayed with `asciinema
-play <filename>` and/or uploaded to asciinema.org with `asciinema upload
-<filename>`. If the `filename` argument is omitted then (after asking for
-confirmation) the resulting asciicast is uploaded to asciinema.org for further
-playback in the browser.
+[asciicast](doc/asciicast-v1.md)) is saved to a local file. It can later be
+replayed with `asciinema play <filename>` and/or uploaded to asciinema.org with
+`asciinema upload <filename>`. If the `filename` argument is omitted then
+(after asking for confirmation) the resulting asciicast is uploaded to
+asciinema.org for further playback in the browser.
 
 `ASCIINEMA_REC=1` is added to recorded process environment variables. This
 can be used by your shell's config file (`.bashrc`, `.zshrc`) to alter the
diff --git a/doc/asciicast-v1.md b/doc/asciicast-v1.md
new file mode 100644
index 0000000..faa3323
--- /dev/null
+++ b/doc/asciicast-v1.md
@@ -0,0 +1,62 @@
+# asciicast file format (version 1)
+
+asciicast file is JSON file containing meta-data like duration or title of the
+recording, and the actual content printed to terminal's stdout during
+recording.
+
+## Attributes
+
+Every asciicast includes the following set of attributes:
+
+* `version` - set to 1,
+* `width` - terminal width (number of columns),
+* `height` - terminal height (number of rows),
+* `duration` - total duration of asciicast as floating point number,
+* `command` - command that was recorded, as given via `-c` option to `rec`,
+* `title` - title of the asciicast, as given via `-t` option to `rec`,
+* `env` - map of environment variables useful for debugging playback problems,
+* `stdout` - array of "frames", see below.
+
+### Frame
+
+Frame represents an event of printing new data to terminal's stdout. It is a 2
+element array containing **delay** and **data**.
+
+**Delay** is the number of seconds that elapsed since the previous frame (or
+since the beginning of the recording in case of the 1st frame) represented as
+a floating point number, with microsecond precision.
+
+**Data** is a string containing the data that was printed to a terminal in a
+given frame. It has to be valid, UTF-8 encoded JSON string as described in
+[JSON RFC section 2.5](http://www.ietf.org/rfc/rfc4627.txt), with all
+non-printable Unicode codepoints encoded as `\uXXXX`.
+
+For example, frame `[5.4321, "foo\rbar\u0007..."]` means there was 5 seconds of
+inactivity between previous printing and printing of `foo\rbar\u0007...`.
+
+## Example asciicast
+
+A very short asciicast may look like this:
+
+    {
+      "version": 1,
+      "width": 80,
+      "height": 24,
+      "duration": 1.515658795,
+      "command": "/bin/zsh",
+      "title": "",
+      "env": {
+        "TERM": "screen-256color",
+        "SHELL": "/bin/zsh"
+      },
+      "stdout": [
+        [
+          0.248848,
+          "\u001b[1m\u001b[3m%\u001b[23m\u001b[1m\u001b[0m"
+        ],
+        [
+          0.001376,
+          "\r\u001b[0m\u001b[23m\u001b[24m\u001b[J\u001b[34"
+        ]
+      ]
+    }