Abstract: In the two-phase flow modeling of reactor systems, the flow pattern is a necessary closure term for establishing constitutive equations. In the reactor system codes, it is necessary to select the thermal-hydraulic calculation correlations based on the flow pattern and consider the possible numerical deviations in the flow pattern transition region. While an accurate flow pattern transition model is a fundamental prerequisite for selecting the constitutive correlations of flow and heat transfer. However, the existing bubbly-to-slug flow pattern transition criteria for small-diameter pipes have certain limitation on applicability or accuracy. Thus, the air-water two-phase co-current upward flow experiment had been conducted in small-diameter pipes, whose IDs were 10, 20 and 30 mm, respectively. The superficial velocities of water and air ranged from 0.25 to 4.0 m/s and 0.008 to 25.5 m/s, respectively. The annular impedance meters were used for void fraction measurement and the high-speed camera was used for flow structure recording. Focused on bubbly-to-slug flow pattern transition, the random forest algorithm was employed to distinguish the bubbly flow and slug flow conditions of transition region with CDF (cumulative distribution function) curves. Based on the experimental data, the flow structure of the conditions that lied in near the bubbly-to-slug flow pattern transition region was divided into two regions with the large bubble region and the liquid slug region, and then the void fraction of each region was obtained. Consequently, the void fraction characteristics of the large bubble region and the liquid slug region was evaluated with the bubble behaviors and possible phase interaction mechanism. The result indicates that, the pipe size is a significant factor affecting the void fraction of bubbly-to-slug flow pattern transition (α_B-S). α_B-S increases with the pipe size, while α_B-S is less influenced with the liquid phase flow rate variation when the pipe size remains constant. At low j_f conditions, the wake entrainment mechanism plays a dominant role on void fraction variation in large bubble and liquid slug regions, while the impact of random collision mechanism is relatively minor. At high j_f conditions, the effect of the turbulence breakup mechanism becomes more pronounced, suppressing the formation and development of large bubbles. While the wake entrainment mechanism still dominates the void fraction variation in large bubble and liquid slug regions. Based on the phase interaction mechanism, a new predictive model for bubbly-to-slug flow pattern transition applied in small-diameter pipes is proposed with the impact of the random collision, wake entrainment and turbulent breakup mechanism. The model predictions are in good agreement with the experimental data, with an absolute average relative error of 5.14% for α_B-S.