A method to recognize the type of woven fabric was developed by employing neural networks. In the analysis, transmission image of the woven fabric was used to obtain the yarn position, while reflection image was used to obtain the ellipsoidal aspect ratio of warp/weft interlacing point. The Learning Vector Quantization algorithm was used as a neural network and trained by applying the aspect ratio as the input value. Three fundamental types of woven fabrics could be successfully recognized by the analysis.