In order to develop a sample of recent-onset marijuana users, the study used the files of the National Household Survey on Drug Abuse (NHSDA), a continuing cross-sectional survey of the drug experiences of U.S. citizens 12 years old and older. Of the 86,021 NHSDA participants during 1995-98, there were 2,628 recent-onset marijuana users; these people (1,866 adolescents and 762 adults) constituted the study sample. For this sample marijuana dependence was assessed by means of seven standardized questions about the person's patterns of marijuana use, such as being unable to reduce use. Given the possibility that the cumulative days of marijuana use might differ across the age-stratified subgroups, the GLM/GEE statistical model adjusted for this characteristics as if it were an exogenous covariate. A MIMIC (multiple indicators, multiple causes) model adapted for binary responses was applied in a more exploratory check on the possibility of age-related differences in reporting of each clinical feature. Findings show that among people who had just begun to use marijuana, clinical features of marijuana dependence occurred twice as often among adolescents compared to adults, even with statistical adjustment for other covariates. MIMIC analyses suggested that adolescent-onset users had somewhat higher levels of marijuana dependence, and there was also evidence of age-associated response bias for some, but not all, clinical features of marijuana dependence. Even with the level of marijuana dependence held constant, adolescent recent-onset users were more likely than adult recent-onset users to report being unable to cut down on their use and to report tolerance. These findings indicate that future research should give more attention to age-related differences in marijuana dependence problems. 4 tables, 2 figures, and 35 references