Many health problems, such as obesity, are caused by determinants at different levels, e.g., intra- and inter-personal, social and environmental levels. In addition, much research data are also collected from clustered sampling units, such as classes, schools, health clubs, communities, states or metropolitan areas. As a result, the commonly used single-level research design and analytical models (e.g., ANOVA) are no longer appropriate for this kind of hierarchical, or multilevel, research problems and data analysis. This is because observational units (e.g., students or participants) nested within experimental units (e.g., schools or communities) are not independent from each other, which violates the basic assumption of most statistical tests that replications of the experiment are independent. Fortunately, the problems can be overcome by employing the recently developed hierarchical linear model (HLM). Derived from the idea of the "slopes-as-outcome" analytical model (Burstein, 1980), HLM is a multilevel analytical model, which takes the hierarchical data structure into consideration in the data analysis. For example, to analyze a two-level data structure (residents and communities) that a community physical activity study employed, the HLM poses two analytical equations: Residents and community. First, the resident-level equation models the relationships among an outcome variable and various resident-level predictors or characteristics. If regression coefficients at this level are different from community to community, an intercepts- and slopes-as-outcomes equation can then be followed at the community level to determine if community factors could explain the variability. As a result, information at both levels could be thoroughly analyzed, eliminating the problems related to the single-level approach. This presentation will describe multilevel research in detail: Its statistical foundation, historical development, and potential applications in physical activity research. Critical issues in designing, conducting and analyzing a multi-level study, such as statistical power, will also be addressed.