In the present work, the flash atomization mechanism has been studied experimentally focussing mostly on the two-phase effluent flashing mode and a mechanistic model on the atomization process has been constructed to predict the size distribution of drops. In experiments, both the internal flow and external flow patterns (that is, the flow pattern before and after the discharge) were considered simultaneously. The flow pattern inside the nozzle was visualized and the void fraction was measured from photographs of the internal flow by using transparent nozzles. External flow (spray) patterns were also photographed to examine the atomization process and to obtain the spray angle data. Drop sizes were measured by using the light scattering method. It was confirmed that the atomization process after the discharge is governed by the internal flow pattern. The internal flow pattern changes from bubbly flow to slug flow, and then to annular flow with increase of the degree of superheat. However, at the injection pressure above 0.3 MPa, the flow pattern is almost the bubbly flow. Bubble formation/growth becomes more vigorous with increase of the superheat, which results in smaller and uniform drop sizes, and a wide spray angle; especially, the size and number density of bubbles at the nozzle exit are the most important parameters which determine the atomization process. To predict the size and number density of bubbles at the nozzle exit, the one-dimensional two-fluid model along with the vapor generation model by Riznic and Ishii has been used. The atomization process is composed of three steps; bubble breakup, formation of ligaments, and breakup of the ligaments by the aerodynamic drag. In the present atomization model, the mechanical energy is assumed to be conserved when the bubbly two-phase jet is converted into a drop (spray) flow through the disintegration process. Adelberg's model has been adopted to describe the disintegration process of ligaments. Mean drop size predicted by the model is in general agreement with the experimental results in the range of the dimensionless superheat above 0.4. It can be concluded that, except for the low superheat range, the mean drop sizes and flow rates(or mass flux) are predictable to a certain extent with the present mechanistic model once the injection condition and the nozzle geometry are given. Further improvement on the ligament breakup model is needed to predict the spray angle and the spatial distribution of drops reasonably.