Cognitive load theory provides guidelines for improving the training of complex cognitive skills and their transfer to new situations. One guideline states that extraneous cognitive load that is irrelevant to the construction of cognitive schemata should be minimised. Experiment 1 (N=26) compares completion problems, conventional problems, and a learner-controlled condition in which learners may choose between problem formats. Completion problems decrease cognitive load during training and have a zero or positive effect on transfer performance. A second guideline states that germane cognitive load that is directly relevant to schema construction should be optimised. In Experiment 2 (N=69) practice schedules of either high or low contextual interference are compared (HCI and LCI). HCI increases cognitive load during training and shows a trend towards higher transfer performance. Experiment 3 (N=87) combines both guidelines in a factorial experiment with the factors problem format (completion vs. conventional) and contextual interference (HCI vs. LCI). It is hypothesised that redirecting attention from extraneous to germane processes will improve training efficiency, i.e. positively affect the balance between cognitive load during training and transfer test performance. In support of this hypothesis, it is found that the completion-HCI group shows highest training efficiency. But transfer test performance for this group is disappointing. The results are discussed in relation to the operationalisation of HCI in combination with completion problems.