In the first part of the paper, criteria are proposed aimed at avoiding the shorcomings often incurred when double dissociations are found in neural nets. Also, some of the available literature on this issue is critically evaluated. In the second part of the paper, several double dissociations in neural nets are demonstrated, which meet the criteria that were previously introduced. These double dissociations are functional in nature and are obtained in distributed nueral nets. The results can be generalized because they are shown not to depend on specific algorithms or types of hardware. In addition, these double dissociations might contribute to clarifying the functional architecture of semantic memory and the existence of domain specific semantic stores.