Missing data are traditionally deleted or replaced by their statistical estimates. These practices, however, are only appropriate for two kinds of missing data: Missing completely at random (MCAR) or missing at random (MAR). MCAR refers to the cases with missing data are the same, or approximately the same, as those without missing; and MAR refers to the cases with missing data are different from those without missing, but the pattern of data missingness is traceable or predictable from other variables in the database. When the missing data is nonignorable, i.e., the pattern of data missingness is non-random or is not predictable from other variables in the database, other statistical methods must be used. Little (1993, 1994, 1995) and others (Glynn, Laird, & Rubin, 1986; Little & Schenker, 1995; Hedeker & Gibbons, 1997) developed a model called "pattern-mixture model" for nonignorable missing data. The model is especially useful for the missing data in repeated measure or longitudinal studies; and it has the following three important analytical characteristics: (a) dividing the subjects into groups depending on their missing data pattern, (b) the missing pattern is a between-subject variable to be used in longitudinal data analysis, and (c) method of analysis must allow subjects to have incomplete data across time. With subject measured at three timepoints, for example, there are 23 possible missing data patterns: (OOO, OOM, OMO, MOO, MMO, OMM, MOM, MMM; where O= Observed and M = Missing); and the data analyses are based on these patterns. This presentation will describe the pattern-mixture model in details, including related assumptions, when and how the model should be applied, and sensitivity analyses on evaluation of the model. Keyword(s): exercise/fitness, measurement/evaluation, research