Visual search is guided by mental representations of target-defining features (attentional templates) that are activated in a preparatory fashion. It remains unknown how many templates can be maintained concurrently, and what kind of costs are associated with multiple-template versus single-template search. Here, we compared the operation of attentional templates during three-colour and single-colour search tasks. Preparatory template activation processes were tracked by measuring N2pc components to task-irrelevant singleton colour probes that appeared in rapid succession during the interval between search displays. These probes attract attention (as indexed by an N2pc) if the corresponding colour template is active at the time when the probe appears. In a three-colour search task where target identity was fully predictable (Experiment 1), only probes that matched the upcoming target colour triggered N2pcs, demonstrating that only a single target template was activated. When three possible colour targets appeared randomly and unpredictably (Experiment 2), probes that matched any of these colours triggered N2pcs, demonstrating that all three templates were activated concurrently. However, relative to a single-colour search task, clear costs emerged in this three-colour task for the attentional guidance towards search targets and for search performance. These costs appear to be linked to inhibitory interactions between simultaneously active search templates. These findings show that while at least three target templates can be maintained in parallel, multiple-template search is still subject to capacity limitations which affect both template-guided attentional guidance and the subsequent selective processing of search targets.