R22 and R407C mass flow rates through straight and coiled adiabatic capillary tubes are analyzed with three ANN models, i.e., the feed forward network with back propagation (BP) algorithm, GA-BP (genetic algorithm optimized BP network) and PSO-BP (BP algorithm combined with particle swarm optimization). The modelled outputs by these ANN methods are compared with experimental data. The results showed that the predicted mass flow rates with the three models of BP, GA-BP and PSO-BP agree quite well with the experimental data with the mean relative error of 3.82 %, 3.14 % and 2.3 % for R22, and 3.17 %, 2.66 % and 2.46 % for R407C, respectively. PSO-BP network is then employed to predict the coiling effect of capillary tubes on the mass flow rate. It is shown that the mass flow rates with coiled diameter between 0.04 m and 0.6 m are about 4 %–13 % lower than that of the straight capillary tube.