This article reports the construction of a model for predicting the rate of dimensional change of fabric composites after wash. The mechanical properties of the fabrics and fused interlinings were tested in FAST system and KES system and the parameters of fabrics and interlinings are analyzed by a objective mathematic method. Through this method, eight principal components are obtained. A BP neural network with a single hidden layer is constructed including eight input nodes, six hidden nodes and one output nodes. During training the network with a back-propagation algorithm, the eight principal components are used as input parameters, while the rate of dimensional change are used as output parameters. All original data are preprocessed and the learning process is successful in achieving a global error minimum. The the rate of dimensional change can be predicted with this training network and there is a good agreement between the predicted and tested values.