This paper presents a back-propagation neural network model to estimate the screening effects of vertical amplitude of surface waves by 3-dimensional in-filled trenches. A total of six input parameters are selected by using genetic algorithm with important parameter analysis and pre-processing transfer functions. The parameters are trench dimension, distance between vibrating foundation and trenches, in-filled material property, etc. The number of hidden layer nodes in the network is determined by Cascade Correlation learning processing. The learning parameters of network are determined by Extended-Delta-Bar-Delta algorithm to regulate learning-rate and momentum constant automatically. The output parameter of network is average vertical amplitude reduction rate. The results show that the neural network model is a very good approach in estimating the screening effect of vertical amplitude of seismic wave.