This paper addresses the multiplicity problem of video events by dividing an event context to its intra-path context and inter-path context. Then, the novel event analysis method is applied for analyzing various interaction events (e.g. smoking, eating, and phoning) happening between hands and faces. This problem is very challenging since there is no prior knowledge (like shape, color, size, and texture) about the hand-hold objects. To address this problem, a novel ratio histogram is first proposed for finding important color bins for locating desired objects through a re-projection technique. Then, a code book method is then used for object tracking and feature extraction. Then, each action is defined as a temporally ordered set of repetitive symbols. In real cases, an event will have various representations under different lighting and weather conditions. To reflect the multiplicity of an event, we divide an event to two parts, i.e., the intra-path context and inter-path context. The first one models the dynamic properties of an event within a path and the second one is to measure the multiplicity of an event between paths. The intra-path context uses Markov chains to capture the repetitions of action primitives. The multiplicity of an event is modeled into the inter-path context using a weighted edit distance. A Bayesian inference scheme is then used to find the best set of event parameters directly from videos and classify events to different clusters. Experimental results show this scheme is effective to detect and analyze different daily events (e.g. smoking, eating, and phoning) even though variant colors and sizes on objects and dressing appearances are handled.