Three event-related potential (ERP) experiments investigated the spatial-verbal dichotomy (emphasized by Baddeley’s model of working memory), selective attention (emphasized by Cowan’s model), and sub-processes in n-back tasks. The studies provide a basis for further clinical research on Alzheimer’s disease. Experiment 1 studied the spatial-verbal dichotomy using n-back tasks with top-down control. It used identical stimuli in both spatial and verbal tasks, and was designed to eliminate interference from perceptual processes. The spatial and verbal tasks differed only in the instructions given before the tasks. Using a model of the task, sub-processes involved in n-back tasks were delineated and analyzed by difference waveforms of ERPs. Domain-specific lateralization was found in a shift sub-process but not in a replacement sub-process. Because information from the irrelevant domain could not be totally excluded by top-down control (which distinguished the spatial and verbal tasks), Experiment 2 recorded information from irrelevant domains. Interactions between irrelevant and relevant domains were found in ERP difference waveforms. Therefore, the results of Experiment 2 suggest that selective attention is unable to exclude interference from the irrelevant domain. Following this conclusion, Experiment 3 adopted a data-driven (bottom-up controlled) methodology, and used different stimuli in spatial and verbal domains in contrast with Experiment 1. In Experiment 3, selective attention was not needed to distinguish spatial tasks from verbal tasks because the spatial and verbal stimuli were different. A different pattern was obtained in Experiment 3; domain-specific lateralization was found only in the replacement sub-process (and not in the shift sub-process). Relationships among n-back task sub-processes, i.e., matching, replacement and shift were determined in advance by logical analyses. From this model, process-specific patterns of domain-specific lateralization, where domain-specific lateralization should exist in the shift sub-process of a conceptual task and in the replacement sub-process of a data-driven task, were predicted under conceptual and data-driven control. The results in Experiment 1 and 3 fitted these predictions, consistent with the proposed logical hypothesis of sub-processes. The similarity between match-specific and domain-specific lateralization suggests avenues for further studies. Using the framework of the model developed here, further investigation could be carried out at the sub-process level using the high temporal resolution of ERPs to decipher the detailed steps of processing that our brains execute in working memory tasks.