The generalizability and concurrent validity of the FLOAT (Formal Lecture Observation Assessment Inventory) were studied. College instructors were rated, using the FLOAT, by their students and by trained observers who viewed videotapes of the classes. The teachers were also asked to evaluate themselves. A previous study of interrater reliability and student-observer agreement indicated that the instrument was reliable and that there were moderate to strong correlations between groups of raters. To study its generalizability, an analysis of variance was performed for the first fifteen FLOAT variables, used by three raters to rate nine instructors upon three occasions. Generalizability coefficients ranged from .66 to .96. Factor analysis of student ratings yielded two factors, one of which explained 61% of the rotated variance. Factor analysis of observer ratings produced similar results; the main factor, task orientation, explained most of the variance. Results of teacher self-ratings indicated a .42 correlation coefficient with student ratings, and .25 with observers. This difference was attributed to the observer's emphasis on category discrimination and to the students' perspective which did not distinguish greatly between teaching process and course content. (GDC)