Two-phase and liquid-liquid flows in microchannels are widely utilized in heat exchangers, microreactors, biological analysis systems, etc. Two-phase flow patterns depend on liquid flow rates, physical properties of the liquids, microchannel size and geometry, and microchannel wall wettability. Though there is a complex interrelation of many adjustable parameters with flow regimes, it is essential to correctly predict two-phase flow patterns in microfluidic devices. The present work adopts two approaches to flow pattern prediction: dimensional analysis and neural networks. Based on the dimensional analysis, the complex We^a·Oh^b has been suggested for universal flow pattern map construction where We is the Weber number, and Oh is the Ohnesorge number of the continuous and dispersed phases. It was shown that the power exponents a = 0.4 and b = 0.6 give the best results for predicting the transition between segmented and continuous flow patterns in microchannels devices of various geometries and liquid properties. The proposed non-dimensional complex allowed us to construct a universal flow pattern map for gas-liquid and liquid-liquid flows in the cases when the ratio of the viscosities of the continuous phase and dispersed phase λ ≤ 1. Another approach was to train a fully-connected neural network to predict flow patterns of two-phase and liquid-liquid flows in microchannels. As a result, 98% accuracy of flow pattern prediction has been achieved.