To enable SEM-based management of a cross-sectional shape or manufacturing process parameters of a semiconductor device pattern to be measured, the association between the shape or parameters and SEM image characteristic quantities effective for estimating the shape or parameters is saved as learning data. The image characteristic quantities are collated with learning data to estimate the shape or to monitor the process parameters. Accuracy and reliability is achievable by calculating three kinds of reliability (reliability of the image characteristic quantities, reliability of estimation engines, and reliability of estimating results) based on the distribution of the image characteristic quantities and then judging whether additional learning is necessary, or selecting and adjusting image characteristic quantities and estimation engine based on the reliability.