To avoid the bias of continuous-time equation model's estimation by numerical filters, the direct identification of the continuous-time model based on the PSO and output error method is proposed, which uses a simplified fourth Runge-Kutta method to the approximate system's output, and used PSO for the numerical optimization to aviod the estimation bias. The consistence of the proposed method is analysed, which shows that the parameters can be estimated asymptotically consistent in an open loop. Simulations show that the proposed method has a good estimation in this case than several continuous-time identification methods such as SRIVC. Finally, the proposed method is used for a good dynamic model of machine with elastic load, which verifies its effectiveness.