An automated pixel and image classification system has been developed to identify texture patterns within images after training with representative texture patterns. Multispectral analysis is applied to ultrasound images to form hyperspaces in which texture patterns are clustered. The clusters in the space are produced using run-length and Markovian texture statistics. Several neural network models can be selected to classify patterns. The system has been implemented in C on a Sun workstation in a window environment. It is highly automated and has potential for clinical applications. Texture patterns found in a series of cardiac ultrasound images of a tumor were used to train the system. The tumor was correcdy identified throughout a series of consecutive, closely-space tomographic images.