Data derived from a year-long study of 43 algebra classes in 13 high schools were collected to assess the differences between high-inference and low-inference measures of teacher behavior as predictors of pupil achievement and attitude change. The high-inference systems rated characteristics such as enthusiasm and clarity while low-inference measures were often frequency counts of objective, specific teaching acts. Student attitudes and achievement were measured at the beginning and end of the year. Achievement measures involved computation, comprehension, and analysis. Attitude measures included liking mathematics versus non-mathematics; mathematics fun versus dull; mathematics easy versus hard; and actual mathematics self concept. Results showed: (1) relationships of low- and high-inference measures to attitude and achievement measures were low; (2) low inference measures produced more significant correlation coefficients with achievement measures than did high-inference measures; (3) high-inference measures produced more significant correlation coefficients with attitude measures than did low-inference measures; and (4) correlations among low-inference measures and achievement measures were greater in magnitude than the correlations among high-inference measures and achievement measures. It was concluded that both types of observation instruments should be used in process-product research. (MH)