The presenc disappoincing situation in our studies of transfer of training has been recently summarised by E. G. Chambers (1) . Experiments on transfer of industrial skills have been few, and are unlikely co succeed unless che skills involved in the tasks are very choroughly analysed. The technique of such analysis is only now becoming developed sufficiently ( 2 ) . Results of recenc experiments using the capstan lathe indicate a high level of cransfer of skills between the components of a task and its totality ( 3 , 4 ) , bur further understanding of transfer is more likely to come from closer study of the nature of skilled performance than from cransfer experiments as such. Recent experiments' on assembly operations show chat skilled performance levels are attained by a selective process by which S chooses with increasing frequency the more successful means of performance. Ss were required co Reach for an object, to Grasp it, to Move it, and to Position it on three pegs. The time for each of these cherbligs was separately recorded on SETAR ( 5 ) . The resulcs indicate: ( a ) That improvements in the Grasp and Positiolz therbligs (which involve more perception) contributed mosc to over-all improvement. Fumbles occurred frequently in these therbligs, and improvement came by a reduction of fumbles and of duracion of fumble-free repetitions. ( b ) Improvements in the cherbligs arose by less frequent occurrence of long rimes and more frequent achievemenc of shorter times, and a consequent "shift" in the distribucion of cimes. How does S improve performance thus? The process must necessarily be one of selection, but the means of selection and its basis are not clear. External knowledge of results, which in other circumstances may be a. basis for selecting particular responses, would not seem to apply in a cask such as chis. It is suggested that the S's criterion of achievement and selection may be based not on time buc on energy expenditure, i.e., the attempts giving rise to shorter cherblig times may be recognised through their involving lers not more activity. All therbligs are performed as a result of muscular contractions achieved by innervation of certain muscle groups. I t is reasonable -