This paper describes a method of dialect discrimination based on differences in slopes of fundamental frequency contours of speech. The fundamental frequency contours were approximated by polygonal lines which were determined using a dynamic programming procedure. The starting frequency and slope of each line were calculated. Mean values, standard deviations and histograms of these parameters were calculated for about 40 seconds of speech spoken by male speakers of standard Japanese and four typical Japanese dialects. Results show that the four dialects and standard Japanese can be discriminated based on the slopes and relative starting frequencies of the approximated lines.