Thus far, network modules have mostly been implemented on computers and electronic hardware with low density interconnections at low speed. Optics potentially offers high density parallel interconnections at high speed. However, the implementation has been limited to single layer machines. As a result, no practical implementation has yet been reported. The lack of suitable materials providing dense synaptic interconnections is the primary reason for the slow progress in the practicality of optical neural network implementation. Photorefractives are considered as suitable materials. However, photorefractives have inherent problems including scattering, sensitivity, low refractive index modulation and fast decay. Furthermore, optical network systems based on photorefractives have low signal-to-noise ratio, have poor stability, and cannot realize error driven learning algorithms, such as the back error propagation. In addition, low diffraction efficiency severely affects cascadability of neural layers. Therefore, multilayer machines using photorefractive crystals are difficult to implement. Physical Optics Corporation (POC) has developed a new dynamic birefringent material for recording polarization holograms with selective enhancement or erasure in real time. Extensive investigation has shown that POC's material offers well-controlled dynamic behavior that is superior to photorefractive crystals, Polaroid DMP-128, Du Pont photopolymer and Kodak's silver halide for neural network interconnection applications.