diff --git a/boxmox/data.py b/boxmox/data.py
index f9b40880051eef8acc886b148487e2786845b05c..e52fdc74284e8bb103e9e4b529ad59f5458ae089 100644
--- a/boxmox/data.py
+++ b/boxmox/data.py
@@ -3,18 +3,20 @@ import sys
 import shutil
 import StringIO
 import csv
+import warnings
 
 import numpy as np
 
 def _mygenfromtxt(f):
     curpos = f.tell()
-    dialect = csv.Sniffer().sniff(f.read(1048576), delimiters=";, ")
-    # for FORTRAN output, we "just" know we have to set this.
-    dialect.skipinitialspace = True
-    f.seek(curpos)
+    try:
+        dialect = csv.Sniffer().sniff(f.read(1048576), delimiters=";, ")
+        f.seek(curpos)
+        spamreader = csv.reader(f, dialect)
+    except:
+        warnings.warn('Could not determine dialect, falling back to default.')
+        spamreader = csv.reader(f, skipinitialspace = True, delimiter=" ")
     # twice as fast as np.genfromtxt(..., names=True)
-#    spamreader = csv.reader(f, skipinitialspace = True, delimiter=" ")
-    spamreader = csv.reader(f, dialect)
     hdr = spamreader.next()
     dat = []
     for row in spamreader: