Solar proton events especially those with high fluxes may cause threat to the spacecrafts and satellites round the orbits near the earth, and may cause damage to the sensitive electronic components on the satellites, therefore, accurate short-term prediction of proton events is very meaningful to assure the safety of the space task and coordinate the instruments aboard the satellites. The current research shows that there exist a considerable correlation between proton events and soft X-ray radiation, so in this paper, based on the 1 ~ 8 A and 0.5 ~ 4A band soft X-ray data from GOES database, and choosing some characteristic parameters for our proton prediction model, a BP neural network was designed and used to predict the peak flux of the proton events, with the network input of soft X-ray data. The test result shows that in most cases the prediction error is less than one order.