<p>Perceptual decisions are shaped by prior information and choice history, yet their distinct computational implementations remain unclear. Here, we jointly model these sources of bias using hierarchical drift diffusion models fitted to behavioral and electroencephalography data from two near-threshold somatosensory detection-confidence experiments with stable (<i>N</i> = 43) and volatile (<i>N</i> = 39) probability environments. Across both datasets, we find that stimulus probability biases the starting point of evidence accumulation, whereas previous choices modulate the drift rate, consistent with a history-dependent accumulation bias. Bayesian model comparison provides decisive evidence for this dissociation across environments. Models with stimulus-dependent accumulation noise are strongly preferred, indicating increased variability in signal trials. Integrating single-trial pre-stimulus beta-band power into the model, we further show that baseline fluctuations mediate the effect of previous choice on drift rate in the stable environment, but not in the volatile environment, whereas no mediation is observed for probability-induced starting point biases. Together, these results establish a mechanistic dissociation between explicit and implicit biases in perceptual decision-making and demonstrate how environmental volatility shapes their neural and computational signatures.</p>