In cortical information processing, neuronal inputs are transformed into sequences of action potentials. However, the neural codes used for embedding information in the spike trains remain unclear. Here a neural network consisting of recurrent inhibitory connections is shown to achieve selective activation and inactivation of neurones very efficiently according to spike timing rather than firing rates, when they are stimulated by periodic spike trains. Oscillatory neural activities serve as an accurate clock for the spike-timing code utilized in the rapid selection of neural activities. These results suggest that differences in spike timing of < 1 ms can be of functional significance in certain neural information processing.