Thee groups consisting of 332, 243 and 280 barley breeding lines (entries) of known selection history were evaluated in 10, 9 and 8 environments, respectively, to determine the relationship between grain yield in low yielding (LYE) or high yielding (HYE) environment, and selection history and type of germplasm. One cycle of selection in LYE produced on average five times more entries outyielding the best check in LYE than selection in HYE. A retrospective analysis indicated that the highest yielding lines in LYE were lower yielding (15%-28%) in HYE when compared with the best check, and by 20% and 38% compared with the best entries in HYE. In contrast, the highest yielding lines in HYE were lower yielding (4%-33%) in LYE when compared with the best check, and by 33% and 40% when compared with the best entries in LYE. The highest yielding lines in LYE did not differ consistently from the highest yielding lines in HYE for a number of morphological and developmental traits including days to heading. This suggests there are many paths to high yield in LYE and that analytical breeding based on individual traits may not be appropriate for variable environments. Only 0.07% of the highest yielding entries in LYE was selected for high yield in HYE conditions confirming previous results indicating that selection for high yield in HYE is an inefficient strategy for improving yield in low yielding conditions. This frequency is 28 times lower than the frequency of high yielding entries in LYE selected from landraces or crosses with landraces in low yielding conditions. The results imply that the most cost-effective strategy for barley breeding in low yielding conditions is to select repeatedly in low yielding conditions and to include adapted germplasm (landraces) in the breeding material.