Classification of Mental Retardation (MR) into severe and profound is based on IQ threshold (<35 and 20% respectively) and on quite generic descriptions of deficits in adaptive behavior. The LAPMER scale (after Level of Activity in Profound/severe Mental Retardation) was developed as a measure of severity through observed behavior in adult patients. The Rasch analysis (RA, in its rating scale model) was adopted as a guide for selection of items, conceptualization of item levels, and validation of the overall instrument. The RA provides estimates on a continuum measure corresponding to the discrete cumulative score. A model prescribes the expected scores on each subject-item interaction. Discrepancies between observed and expected scores allow diagnostic procedures on coherence (fit) of both subjects and items. The final version included 8 items: Feeding, Sphincters, Communication, Manipulation, Dressing, Locomotion, Spatial Orientation and Praxiae, scored 0/1 or 0/1/2 (cumulative range for the total set of items was 0-13) the higher the score, the better the performance. The test can be administered in 15 minutes through observation or inquiry from proxies and personnel. A psychologist rated 146 permanent hosts of a large Institute for mentally retarded adults (51 profound and 95 severe, 91 male, age 18-63, median 36). Median score was 6/13, IQR 1-9, range 0-12, 19% of cases scored 0. Cronbach a for internal consistency was 0.90. Fifty-seven patients were also independently scored by another psychologist. Between-rater Cohen's k reliability index ranged from 0.77-0.96 across items. Median raw scores were 1 and 8 in profound and severe cases, respectively (p<0.001). Rasch person reliability coefficient, a 0 to 1 index of internal consistency analogous to Crohnbach a, was 0.92. For each item the standardized differences between observed and model-expected scores (residuals) were c2 tested (a level 0.05) across sub-groups of patients. These were: profound vs. severe cases, and classes of motor impairment (tetra-,hemi-,para-plegic and unimpaired), matched for overall ability measure. For 6 items some residuals were found to be statistically significant. Absolute differences ranged from 0 to 0.7 raw score points, with no systematic patterns. Gender, age group and rater did not bias the measure. Residuals did not correlate meaningfully across pairs of items (r<(0.5)), further supporting the unidimensionality of the measure. The scale seems a valid tool for classification of adult severe and profound MR cases.