Use proc standard. This will give you the standard normal distribution of your variables. Remove the variables outside the range of -3 to 3.
This only treats univariate outliers. For multivariate outliers look at Mahalanobis distances. This can be done using proc princomp using std option in the statement. Other method is to do regress any unwanted variable (say row number) on the variables of interest (IVs) & calculate INFLUENCE statistic. This can be done in proc reg.
proc reg data=;
model row_num= a b c/ influence;
This will give you a variable called leverage denoted in output as Hat diag H. Mahalanobis distance= MD=(n-1) (hat_diag_h-1/n). MD has a chi-sq distribution with degree of freedom =# of variables. This will give you the cut off beyond which you can label the observations as outliers.