A method for optimizing MR imaging pulse sequence parameters in a statistical framework is presented. Parameters are defined to be optimal when the resulting scalar images yield optimal image segmentations using Bayesian pixel classification. Thus, Bayes risk is used as the objective function to minimize. Approximations are made to give a tractable solution in a four-step procedure. A sample calculation is carried out to determine the optimal TR and flip angle for scalar SPGR imaging of the brain. Overall, this paper gives a new approach to optimize MRI pulse sequences for the specific objective of improved image segmentation.