Bayesian SZNet predicts spectroscopic redshift through use of a Bayesian convolutional network. It uses Monte Carlo dropout to associate predictions with predictive uncertainties, allowing the user to determine unusual or problematic spectra for visual inspection and thresholding to balance between the number of incorrect redshift predictions and coverage.