In the visual search literature, an attentional template refers to a mental representation of target features stored in visual working memory. The dual-state model (Olivers et al., 2011) argues that visual working memory can only be focused on a single representation at a time, and therefore, we can only search for one target at a time. Consequently, to detect two potential targets, we must rapidly alternate our attention between the two corresponding representations. By contrast, the resource hypothesis (Huynh Cong & Kerzel, 2021) posits that we can simultaneously activate more than one representation for visual search, although this requires sharing the available resources. In three pre-registered experiments, we compared the predictions from the state-based model and the resource hypothesis in a combined search and memory task. The results suggest that participants allocate their resources based on the reliability of the retro-cue and therefore favor the resource hypothesis. In contrast, the results are less consistent with the state-based model, which predicts that the cost in terms of response time and accuracy should remain identical regardless of the retro-cue’s reliability.