近年来 ,前馈神经网络广泛地应用在 Logit回归作为标准统计方法的分析领域 .但却很少作它们之间的直接比较 ,本文是 Logit回归和前馈神经网络“比较研究”的一个尝试 ,说明了一些理论结果和特性 ,讨论了在它们的应用中碰到的一些实际问题 ,还进一步用分析的和模拟的两种方法研究了一些重要的渐近概念、过分拟合以及模型选择等问题 ,最后讨论并给出一些结论 In recent years, feedforward neural networks are widely used in the field of analysis of Logit regression as a standard statistical method, but seldom for direct comparison between them. This paper is an attempt to “comparatively study” Logit regression and feedforward neural networks, Some theoretical results and characteristics are described, some practical problems encountered in their application are discussed, and some important asymptotic concepts, overfitting and model selection are further studied by using two analytical and simulation methods Finally, we discuss and give some conclusions