This paper describes a new method for assigning letter grades to students based on their raw scores, which I call Multi-Curve Grading (MCG). The intuition behind the method is that a class can be composed of several different subgroups, each of which should be assigned a different grade. In this, the method quantifies and builds upon the Distribution Gap grading method. I model the raw scores as coming from a Normal Mixture, with each component of the mixture corresponding to a different letter grade. I estimate this model using Gibbs Sampler. Based on this model, I calculate the probability that each student’s raw score corresponds to each possible letter grade. The grader’s degree of leniency is used to specify his loss function and thus to assign the most optimal letter grades. I compare Multi-Curve Grading to other common grading methods, such as the Standard Deviation method. It appears to assign grades better than these other methods.