The critical heat flux (CHF) sets the upper limit of efficient heat removal for pool boiling. Microstructures fabricated on a heat transfer substrate can effectively increase the limit of heat removal and delay the boiling crisis. The exact physics mechanisms behind microstructure enhancement still remain ambiguous and CHF prediction on microstructured surfaces is not well resolved even if numerous related studies and experiments have been performed. In this study, the deep belief network (DBN) is proposed to predict CHF and study parametric trends of CHF by collecting relevant CHF datasets from published papers. Performance comparisons with other four common machine learning techniques and three modified Zuber models accounting for the effects of microstructures are conducted for exploring complicated and nonlinear relation between CHF and microstructures. Different from the training process of other regression modelling problems, a special model convergence, which is defined in Subsection 3.1, is required to be incorporated into the CHF model of DBN for exhibiting accurate parametric trends of CHF and improving the prediction accuracy. Numerical results demonstrate that DBN can achieve the best performance of CHF prediction in terms of prediction accuracy. The presented methodology provides new insights for CHF modelling in pool boiling enhanced by microstructures.