diff --git a/python/calcratio.py b/python/calcratio.py
new file mode 100755
index 00000000..0c52a16f
--- /dev/null
+++ b/python/calcratio.py
@@ -0,0 +1,109 @@
+#!/usr/bin/env python
+#
+# This script takes directories which contain the hpack-test-case json
+# files, and calculates the compression ratio in each file and outputs
+# the result in table formatted in rst.
+#
+# The each directory contains the result of various HPACK compressor.
+#
+# The table is laid out so that we can see that how input header set
+# in one json file is compressed in each compressor.
+#
+# For hpack-test-case, see https://github.com/Jxck/hpack-test-case
+#
+import sys, json, os, re
+
+class Stat:
+
+    def __init__(self, complen, srclen):
+        self.complen = complen
+        self.srclen = srclen
+
+def compute_stat(jsdata):
+    complen = 0
+    srclen = 0
+    for item in jsdata['cases']:
+        complen += len(item['wire']) // 2
+        srclen += \
+                  sum([len(list(x.keys())[0]) + len(list(x.values())[0]) \
+                       for x in item['headers']])
+    return Stat(complen, srclen)
+
+def format_result(r):
+    return '{:.02f} ({}/{}) '.format(r.complen/r.srclen, r.complen, r.srclen)
+
+if __name__ == '__main__':
+    entries = [(os.path.basename(re.sub(r'/+$', '', p)), p) \
+               for p in sys.argv[1:]]
+    maxnamelen = 0
+    maxstorynamelen = 0
+    res = {}
+
+    stories = set()
+    for name, ent in entries:
+        files = [p for p in os.listdir(ent) if p.endswith('.json')]
+        res[name] = {}
+        maxnamelen = max(maxnamelen, len(name))
+        for fn in files:
+            stories.add(fn)
+            maxstorynamelen = max(maxstorynamelen, len(fn))
+            with open(os.path.join(ent, fn)) as f:
+                input = f.read()
+            rv = compute_stat(json.loads(input))
+            res[name][fn] = rv
+            maxnamelen = max(maxnamelen, len(format_result(rv)))
+    stories = list(stories)
+    stories.sort()
+
+    storynameformat = '{{:{}}} '.format(maxstorynamelen)
+    nameformat = '{{:{}}} '.format(maxnamelen)
+
+
+    sys.stdout.write('''\
+hpack-test-case compression ratio
+=================================
+
+The each cell has ``X (Y/Z)`` format:
+
+X
+  Y / Z
+Y
+  number of bytes after compression
+Z
+  number of bytes before compression
+
+''')
+
+    def write_border():
+        sys.stdout.write('='*maxstorynamelen)
+        sys.stdout.write(' ')
+        for _ in entries:
+            sys.stdout.write('='*maxnamelen)
+            sys.stdout.write(' ')
+        sys.stdout.write('\n')
+
+    write_border()
+
+    sys.stdout.write(storynameformat.format('story'))
+    for name, _ in entries:
+        sys.stdout.write(nameformat.format(name))
+    sys.stdout.write('\n')
+
+    write_border()
+
+    for story in stories:
+        sys.stdout.write(storynameformat.format(story))
+        srclen = -1
+        for name, _ in entries:
+            stats = res[name]
+            if story not in stats:
+                sys.stdout.write(nameformat.format('N/A'))
+                continue
+            if srclen == -1:
+                srclen = stats[story].srclen
+            elif srclen != stats[story].srclen:
+                raise Exception('Bad srclen')
+            sys.stdout.write(nameformat.format(format_result(stats[story])))
+        sys.stdout.write('\n')
+
+    write_border()