Hi, i just committed a fix for issue #135, which is about timedelta rounding. The problem was in drange: before the fix it used numpy.arange which resulted in rounding issues, the numpy doc says about arange: "When using a non-integer step, such as 0.1, the results will often not be consistent. It is better to use linspace for these cases." So i tried to create a fix involving linspace, you can see it here: https://github.com/faucon/matplotlib/commit/06195c35f4348f37002e850d1cee992d07f5a29c it works (and is the only way i found to avoid the roundig issues) but it feels somehow "hackish". An alternative would be the following implementation, which is quite readable but much slower: def drange(dstart, dend, delta): # its very slow """ Return a date range as float Gregorian ordinals. *dstart* and *dend* are :class:`datetime` instances. *delta* is a :class:`datetime.timedelta` instance. """ dloop = dstart datelist = [] while dloop < dend: datelist.append(dloop) dloop += delta return dates.date2num(datelist) i will try to add a test the next days. Then i would create a pull-request if the fix looks reasonable to you. thanks, Maximilian