The overall heat transfer rate due to boiling of an impinging subcooled liquid jet is attributed to several simultaneous mechanisms including liquid and vapor phase convection, quenching (transient convection) and evaporation. In the present research, the Rensselaer-Polytechnic Institute (RPI) wall-boiling model is employed to carry out computational simulations on subcooled turbulent jet impingement boiling of water in a confined and submerged configuration. The effects of the controlling parameters, viz. heater-nozzle size ratios (w(H)/w(N)), degree of surface superheats (Delta T-sat,), jet-Reynolds numbers (Re) are studied on the partitioning of the total heat flux into liquid phase convection, quenching (transient convection) and evaporation. The effects of these controlling parameters were studied in the ranges 1 <= w(H)/w(N) <= 11, -5 degrees C <= Delta T-sat <= 25 degrees C (heat fluxes upto 200 W/cm(2)) and Re = 2500 and 3750, for isothermal and isoflux heaters. The present computational approach was validated by comparison of the local temperature distribution on the heater, and boiling curves against experimental data in literature for axisymmetric and slot jets. The liquid phase convective and quenching components of the total heat flux were found to be larger for relatively lower values of w(H)/w(N), implying smaller heaters. On the contrary, the evaporative counterpart was found to be smaller for relatively smaller heaters upto about w(H)/w(N) = 3, beyond which the change was negligible with increase in w(H)/w(N). Beyond a threshold surface superheat, the heat flux due to quenching was found to be the largest contributor to the total heat flux; and this threshold value of surface superheat decreased almost exponentially with increase in heater size. It was also observed that irrespective of the heater size, the liquid phase convective component of the total heat flux for Re = 3750 was consistently larger as compared to Re = 2500, while the quenching and evaporative components of the total heat flux were nearly equal for both Reynolds numbers. An artificial neural network (ANN) was trained and tested with data sets generated from the present research, and ready-to-use weights are presented. (C) 2012 Elsevier Ltd. All rights reserved.