The problem studied is that of utilizing measured concomitant information in sequential trials for the comparison of two treatments. The analysis is based on the treatment 1 minus treatment 2 response differencey=y1—y2and the corresponding concomitant observation differencex=x1—x2, wherexandyare assumed to be distributed bivariate normally. A simple asymptotically valid test procedure and its performance in Monte Carlo experiments are described. It is shown that appreciable covariance economy may be expected when ρ, the correlation coefficient between the (x, y) values, is high since average sample numbers are reduced by the factor (1 — ρ2) from those obtained in the absence of covariance.