Hardware inaccuracy and imprecision are important factors when implementing neural algorithms. This text presents a study of synaptic weight noise as a typical fault model for analogue VLSI realizations of MLP neural networks, and examines the implications for learning and network performance.