文章研究了在构造复杂地区（如高陡构造带、断层发育等），地震记录上储层的信噪比低的情况下，用人工神经网络方法预测气水边界。对影响预测效果的某些因素作了试验，其中包括地震特征参数的选择、计算时窗大小、数据圆滑方式、学习样本与预测区域的选择等。文中还给出了已知剖面的预测结果，并被钻井证实是成功的。 The article studies the prediction of gas-water boundary by means of artificial neural network when the signal-to-noise ratio of reservoirs in seismic records is low under the condition of complicated structures (such as steep tectonic belt, fault development). Experiments have been made on some factors that affect the prediction performance, including the selection of seismic characteristic parameters, the calculation of time window size, the data smoothing method, the selection of learning samples and prediction regions. The prediction results of known sections are also given and proved to be successful by drilling.