We motivate score matching as an alternative to maximum likelihood estimation, and consider its advantages and shortcomings.