Spiking neural networks (SNNs) are more powerful than their non-spiking predecessors as they can encode temporal information in their signals, but they also need different and biologically more plausible rules for synaptic plasticity. In this paper, an unsupervised learning algorithm is introduced for a spiking neural network. The algorithm is based on the Hebbian rule, with the addition of the principle of adapting a layer activation level so as to guarantee frequent firings of neurons for each layer. The proposed algorithm has been illustrated with experimental results related to the detection of temporal patterns in an input stream.