To provide a pattern classification device capable of continuing correct calculations even if the status values of pixels belonging to a certain area are distributed in such a way as to concentrate at one point, with regard to a classification problem using a test of hypothesis.

The pattern classification device includes a conversion device for converting multimedia data into d-dimensional vectors (prototypes), a storage device for storing the d-dimensional prototypes, a dividing device for dividing, according to an assembly of N d-dimensional prototypes stored in the storage device, the assembly of prototypes into a plurality of classes, and a class parameter calculation device for calculating the average vectors and covariance matrices of the classes provided by the dividing device. The pattern classification device specifies the class of one d-dimensional test vector using the class parameter provided by the class parameter calculating means. The class parameter calculation device is provided with a means for calculating the lower limit of the covariance matrices.