One of the major activities in Natural Language Processing is determining a word’s part of speech (POS) tag. In this research we focus on improving the accuracy of Persian part of speech tagging by applying post processing heuristic rules. To evaluate the effects of those rules we use Bijankhan tagged corpus and for tagging, Maximum Likelihood Estimation (MLE) approach is selected because of its simplicity and the ease of implementation. Furthermore, we have studied the effect of size of training on the accuracy of the MLE method. The experimental results show that the heuristic rules improve the accuracy especially for the unknown words.