Proportional-integral-derivative (PID) controllers are widely used in flight control systems. However, it is often very cumbersome to adjust the parameters of a PID controller. In this study, we use Probabilistic Inference for Learning Control (PILCO) to optimize the parameters of a PID controller. As the first step, we develop a probabilistic dynamics model of the flight control system using input and output data. Next, the existing PID controller is evaluated using the policy evaluation method. Finally, the evaluated PID controller is optimized by policy update. The sampling frequency of the system is 100 Hz and the data acquisition time per round is 8 s. The optimized PID controller can achieve stable control post 10 rounds of offline training. Through PILCO optimization, the flight attitude simulator performed robustly in a fixed-point experiment, indicating that PILCO has tremendous potential in solving nonlinear control and parameter optimization problems.