A stimulus-sampling model of recognition memory is presented that predicts both proactive and retroactive interference. To test the predictions of the model, a recognition memory experiment was carried out using a standard proactive-retroactive design with a forced-choice task. Both accuracy and latency were measured. The data showed, as predicted, equal proactive and retroactive effects on accuracy, providing solid support for the model. The interference effects are interpreted in terms of the model as arising from an increase in indirect marking, the marking of shared stimulus elements in words other than the study word. The model has two parameters representing the rate of indirect marking for high-frequency and low-frequency words and two parameters reflecting the efficiency of direct marking. The latency results completely paralleled the accuracy findings, showing proactive and retroactive effects. A possible extension of the model to handle latencies is considered.