Using the concept of the shrinkage technique [J.F. Thompson, Some shrinkage techniques for estimating the mean, J. Am. Stat. Assoc. 63 (1968), pp. 113–122], Bouza [Robustness of shrunken predictors in stratified populations, Biom. J. 36 (1994), pp. 95–102] studied the robustness of the shrunken predictors measured in terms of bias and mean square error in stratified sampling under super-population model ξ (0, 1:1). He showed that the robustness depends on stratified balanced sample and the difference between the common slope (β ) and uncommon slopes (βh) to strata. In the present paper, a new set of predictors is proposed under the model ξ (0, 1:1) in stratified sampling. The proposed predictors are compared with that due to Bouza [Robustness of shrunken predictors in stratified populations, Biom. J. 36 (1994), pp. 95–102] and it has been found more efficient under certain conditions. The robustness of the proposed predictors is also studied on the same lines as discussed by Bouza [Robustness of shrunken predictors in stratified populations, Biom. J. 36 (1994), pp. 95–102], and it has been found that robustness depends upon the difference between β and βh. A limited simulation study also showed that proposed predictors are more efficient than Bouza's predictors.