We applied the conventional multivariate analysis to detect a mismatch field in individual subjects’ evoked magnetoencephalographic (MEG) responses. Multichannel MEG signals at time points spanning a certain duration were formed into measurement vector. The problem was then to determine the significance of the difference between two mean vectors. While the conventional multivariate analysis is well known, it has not been used for the above purpose. Although other new nonparametric methods have been proposed and have proved promising for similar purposes, parametric methods will continue to be used and it is necessary to examine the behaviors of the conventional method when applied to mismatch field detection. Simulation study was conducted using known signal sources in a brain model. Simulated MEG signals were obtained by ‘forward calculation’ and they were converted to measurement vector. The significance of the difference between mean responses was calculated while the parameter values controlling the signal source and the vectorization conditions were changed. The simulation results showed dependence of the detection power on parameter values, and demonstrated that multivariate analysis works well to detect a mismatch field in individual subjects. We also applied the method to human MEG responses to omission of a tone in the musical scale and showed that the brain responses to this stimulus were diverse among subjects.