The objective of this study is to model the lateral interactions between motorized vehicles (MVs) and non-motorized vehicles (NMVs) in mixed traffic. Road user trajectories from two locations in China are extracted using computer vision techniques. The critical lateral distance (the shortest lateral distance to initiate avoidance maneuvers) is used as the lateral interaction indicator. Lateral interactions are modelled using the parametric accelerated failure time (AFT) duration model with a Weibull distribution, and the unobserved heterogeneity is considered using gamma frailty. The results show that interaction probabilities increase at higher MV speeds or NMV-MV speed differences and decrease with the NMV or MV yaw rates. The critical lateral distances when NMV ride in the MV lanes are shorter than those in the NMV lane. Moreover, bikes have higher interaction probabilities than e-bikes. These findings give insights into lateral interaction behaviours in mixed traffic and support better designs of such facilities.