An evaluation method for calculating an identification signal by using PRML, comprises a table storing true patterns and ideal signals thereof, false patterns corresponding to the true patterns and ideal signals thereof, and Euclidean distance between the true pattern and the false pattern, a calculating unit, when a recording signal synchronized with an identification signal coincides with any pattern of the table, calculates a distance between Euclidean distance between the ideal signal of the true pattern and a reproduction signal and Euclidean distance between the ideal signal of the false pattern and a reproduction signal, and an evaluation unit which evaluates an identification signal by using an average and standard deviation of difference. A table for patterns is created for a likely mistaken pattern, whereby calculating an evaluation value with high precision, of a signal quality, with a small amount of calculation.