This paper presents a summary of four research projects presented at MICRONEURO 94, covering a variety of different hardware implementations of Artificial Neural Networks. The first two works describe optical and optoelectronic implementations. A combination of optics and electronics is described in the first work. An optical input plane for a neural net has been built so thatwhole images with tens of thousands of pixels can be entered into a network in parallel. In the second work, an all-optical network is presented, where not only the communication, but also the calculations, are done optically by using optically nonlinear materials. The third work addresses the issue of on-chip learning in analog implementations, by comparing the required precision for different learning schemes. It is observed that traditional algorithms such as back-propagation require a high resolution in the computation. The fourth work describes a digital VLSI circuit implementing a self-organizing feature map, an unsupervised learning technique. Also in this example one of the major problems is the resolution of the computation.