The neural systems sustaining object naming were examined using the activation likelihood estimation (ALE) meta-analysis approach on the results of 16 previously published studies. The activation task in each study required subjects to name pictures of objects or animals, but the baseline tasks varied. Separate meta-analyses were carried out on studies that used: (1) high-level baselines to control for speech processing and visual input; and (2) low-level baselines that did not control for speech or complex visual processing. The results of the two meta-analyses were then compared directly, revealing a double dissociation in the activation pattern for studies using high and low baselines. To interpret the differential activations, we report two new functional imaging experiments. The aim of the first was to characterize activation differences associated with visual stimuli that are typically used in baseline conditions (complex visual features, simple structures, or fixation). The aim of the second was to classify object-naming regions in terms of whether they were engaged preferentially by semantic or phonological processes. The results reveal a remarkably precise correspondence between the areas identified by the meta-analyses as affected differentially by baseline and the areas that are affected differentially by non-object structure, semantics or phonology. As expected, high-level baselines reduced object-naming activation in areas associated with the processing of complex visual features and speech production. In addition, high-level baselines increased sensitivity to activation in areas associated with semantic processing, visual-speech integration and response selection. For example, activation in the anterior temporal areas that neuropsychological studies have associated with semantic processing was more strongly activated in the context of high-level baselines. These results therefore have implications for understanding the convergence of functional imaging and neuropsychological findings.