Picture-word interference refers to the fact that when a picture (i.e., line drawing) is presented with a word superimposed, picture naming latency is longer than when the picture is presented alone. In addition, naming latency will be further prolonged whenever the word and the picture are members of the same semantic category. This semantic interference effect was investigated in a series of studies in order to develop an appropriate model of the semantic processes involved in picture-word interference. In Experiments 1 and 2, it was determined that the associative strength between the word and the picture is unimportant in the picture-word interference task. In Experiment 3, it was demonstrated that the category typicality of the word and the picture is also unimportant in this task. These results suggest that the semantic processes in picture-word interference would not be well described by a semantic network model. This conclusion was reinforced by Experiment 4, in which it was found that the imageability of a word is a highly important factor in the picture-word interference task. The present set of results suggests that any model of the processes involved in picture-word interference must have at its core the notion of the word's "relevance" to the task of naming the presented picture.