The neural formation of mutually exclusive decisions is assumed to reflect noisy evidence accumulation in favor of one or another response. This process is formally described by sequential sampling models from which the drift diffusion model has been empirically well supported. In order to adapt to varying time horizons, it has been proposed that decision boundaries may dynamically change in the time-scale of hundreds of milliseconds, yet such neural signals have not been shown in humans. By measuring the lateralization of EEG beta band power desynchronization towards the motorcortex triggering a response in a speeded dual choice task in a large sample of healthy human individuals, we derive an online index of this decision formation. We find that this beta signal complies to various predictions of accumulator class decision models, including pre-lateralization before the evidence accumulation begins which determines speed and accuracy of subsequent responses. Collapsing decision boundaries terminate decision making under time pressure and are supported by the beta signal. Distracting information is represented in beta band as well, inducing incorrect responses.This indicates that mutually exclusive decisions adhere to fundamental principles of sequential sampling models and confirms lateralized beta power as a valid online marker of decision formation that can be employed in various tasks and predicts errors and reaction times before stimuli appear on screen.