The objectives of this study were (1) to compare smoking between recovered and non-recovered patients with borderline personality disorder (BPD) over the course of 18 years and (2) to assess baseline predictors of tobacco use in patients with BPD. A total of 264 borderline patients were interviewed concerning their smoking history beginning at the 6-year follow-up wave in a longitudinal study of the course of BPD (McLean Study of Adult Development) and re-interviewed at 2-year intervals over the next 18 years. Initial data collection of the larger study happened between June 1992 and December 1995, and the and the Revised Diagnostic Interview for Borderlines (DIB-R) were used as the diagnostic instruments for BPD. Recovered patients had a 48% lower prevalence of smoking than non-recovered patients at 6-year follow-up (a significant difference;  = .01). Also, the rate of decline in smoking for the recovered group was 68% and was significantly faster ( = .008) than for the non-recovered group over the subsequent 18 years. Alcohol abuse or dependence (relative risk [RR] = 1.22; 95% CI, 1.06-1.40;  = .005), lower levels of education (RR = 1.28; 95% CI, 1.15-1.42;  < .001), and higher levels of the defense mechanism of denial (RR = 1.08; 95% CI, 1.03-1.13;  = .002) were significant predictors of smoking in borderline patients in multivariate analyses. Taken together, the results of this study suggest that recovery status was an important element in the prevalence of smoking among borderline patients over time. They also suggest that smoking was predicted by 3 factors: prior psychopathology, demographics, and psychological maturity.