Grid computing is a recently developed technology. Although the developmental tools and techniques for the grid have been extensively studied, grid reliability analysis is not easy because of its complexity. This paper is the first one that presents a hierarchical model for the grid service reliability analysis and evaluation. The hierarchical modeling is mapped to the physical and logical architecture of the grid service system and makes the evaluation and calculation tractable by identifying the independence among layers. Various types of failures are interleaved in the grid computing environment, such as blocking failures, time-out failures, matchmaking failures, network failures, program failures, and resource failures. This paper investigates all of them to achieve a complete picture about grid service reliability. Markov models, queuing theory, and graph theory are mainly used to model, evaluate, and analyze the grid service reliability. Numerical examples are illustrated