Beforehand, the device characteristic patterns of each critical dimension SEM are measured, a sectional shape of an object to undergo dimension measurement is presumed by a model base library (MBL) matching system, dimension measurements are carried out by generating signal waveforms through SEM simulation by inputting the presumed sectional shapes and the device characteristic parameters, and differences in the dimension measurement results are registered as machine differences. In actual measurements, from the dimension measurement results in each critical dimension SEM, machine differences are corrected by subtracting the registered machine differences. Furthermore, changes in critical dimension SEM's over time are monitored by periodically measuring the above-mentioned device characteristic parameters and predicting the above-mentioned dimension measurement results. According to the present invention, actual measurements of machine differences, which require considerable time and effort, are unnecessary. In addition, the influence of changes in samples over time, which is problematic in monitoring changes in devices over time, can be eliminated.