According to one embodiment, a pattern recognition method includes calculating similarities of the input pattern with respect to respective categories, converting the calculated similarities of the input pattern with respect to the respective categories into first evaluation values based on a first table which indicates a relationship between similarities for respective categories and first evaluation values, calculating second evaluation values based on the calculated first evaluation values for the respective categories and prior probabilities for the respective categories stored in a second table indicating prior probabilities of the respective categories, and selecting a category corresponding to a maximum value of the calculated second evaluation values.