一本文提出了一种用前馈神经网络对带噪声样本的去噪声建模的实验方法，能获得合适的网络模型，并具有较好的去噪声能力。实验对比了BP网络和RBF网络在去噪声能力上的差别，结果表明，RBF网络去噪声能力优于BP网络。这一结论已被用于为半导体生产工艺控制参数优化的去噪声建模中。 In this paper, an experimental method for modeling de-noising of noisy samples by using feedforward neural network is proposed. An appropriate network model can be obtained with good denoising ability. The differences between the BP network and the RBF network in the de-noising capability are experimentally compared. The results show that the de-noising capability of the RBF network is better than that of the BP network. This conclusion has been used in the de-noising modeling of process control parameters optimization for semiconductor manufacturing.