Multimodal spellers combining visual and auditory stimulation have recently gained more attention in ERP-based Brain-Computer Interfaces (BCIs). Most studies found an improved efficiency compared to unimodal paradigms while few have explored the effect of the visual-to-auditory delays on the spelling performance. Here, we study five conditions with different visual-to-auditory delays, in order to find the paradigm that provides the best overall BCI performance. We compared the temporal and spatial binary classification accuracy as well as the grand-averaged classification accuracies over repetitions. Results show that long delays may cause better performance in early time intervals corresponding to negative ERP components, but better overall performance is achieved with short visual-to-auditory delays.