Why is there a difference between BER curves for QPSK modem for theoretical & Monte Carlo simulation?

Hi, I have recently built a QPSK modem in Simulink environment. The simulation runs OK and I get the waterfall curve as usual. However, there is a slight differece between the theoretical curve (i.e. obtained from erfc function) and Monte Carlo simulation that is based on counting the number of wrong bits after passing the signal through a noisy channel. The following RAR file contains a snapshot as well as the modelled model file. http://www.4shared.com/rar/vNR2cGYJ/regQPSKmodem.html

So many thanks to you in advance,

Mazen Shanawani