The major function of prefrontal cortex (PFC) is known as working memory, which retains relevant information on-line. Rao et al. (1997) found neurons contributing to both object and spatial working memory. However, its mechanism is still unknown. In this study, we propose a neural network model of working memory in order to shed light on the mechanism. Our model has two input streams and can cope with the task in which two kinds of information have to be retained at the same time. We simulated some physiological results with this model. As a result, we could simulate temporal activity patterns of the neuron responding to both object and location information as shown in Rao et al. (1997). And we considered domain-specificity by constructing three architectures of neural networks and the physiological results could be simulated best by no-domain specificity model. This result suggests that there should be no domain-specificity in PF working memory.