Multiple outlying observations are frequently encountered in applied studies in business and economics. Several multiple outlier tests exist, but little evidence is available on their relative power against alternative types of outliers and influence. This paper analytically and numerically compares the sensitivity of these statistics to particular forms of data contamination. It also proposes new statistics. Practical issues associated with the application of these tests are illustrated by re-examining recent evidence of an R and D-productivity slowdown. New data from Griliches and Lichtenberg (1984) are used to show that the low rates of return to federal and private R and D found in other studies may be due to the presence of outliers.