This study estimates the prices and variances of collateralized mortgage obligations (CMOs) using Monte Carlo simulation, and then examines the relation of structure and risk of CMOs. The simulation results show that Z-tranch tends to reduce the risk of sequential-pay tranch. This reduction is positively related to the amount of Z-tranch and the term of Z-tranch. The effect of floating-rate tranch, however, is unclear. Final, in estimating the risk of CMOs, the traditional method of average life produces large bias. The next is the OLS model. The neural network estimation results in very little bias, and is the best tool for CMOs risk measurement.