This study investigates the effects of four computer-based training strategies for geometrical problem solving in the domain of computer numerically controlled (CNC) machinery programming. The strategies compared were low- and high-variability conventional conditions (solving conventional problems followed by worked examples) and low- and high-variability worked conditions (studying worked examples). Results showed that students who studied worked examples gained more from high-variability examples, invested less time and mental effort in practice, and achieved better and less effort-demanding transfer performance than students who first attempted to solve conventional problems and then studied worked examples. The study highlights the importance of cognitive load in learning complex cognitive tasks. Cognitive load is a multidimensional construct that includes task environment characteristics, subject characteristics, and interactions between them. It can be measured through mental effort, which reflects the amount of controlled processing engaged by the individual. The study found that conventional problem-solving methods, which often involve means-ends analysis, can impose high cognitive load due to extraneous processes not directly relevant to learning. The study also emphasizes the role of practice-problem variability in schema acquisition and transfer. Training with worked examples, especially high-variability ones, was found to be more effective in facilitating schema acquisition and transfer performance. This is because worked examples provide stereotyped solutions and help in rule automation, which are essential for problem-solving transfer. The study used a cognitive-load approach to measure mental effort through rating scales and heart-rate variability spectral analysis. It found that students in the worked conditions perceived less mental effort and achieved better transfer performance than those in conventional conditions. The results suggest that worked examples are more efficient in reducing cognitive load and enhancing transfer performance. The study concludes that effective instructional strategies should focus on reducing cognitive load by directing attention to aspects that facilitate schema acquisition. Worked examples, especially with high variability, are particularly effective in this regard. The findings support the cognitive-load theory and suggest that future research should explore the relationship between cognitive load and learning processes in more depth.This study investigates the effects of four computer-based training strategies for geometrical problem solving in the domain of computer numerically controlled (CNC) machinery programming. The strategies compared were low- and high-variability conventional conditions (solving conventional problems followed by worked examples) and low- and high-variability worked conditions (studying worked examples). Results showed that students who studied worked examples gained more from high-variability examples, invested less time and mental effort in practice, and achieved better and less effort-demanding transfer performance than students who first attempted to solve conventional problems and then studied worked examples. The study highlights the importance of cognitive load in learning complex cognitive tasks. Cognitive load is a multidimensional construct that includes task environment characteristics, subject characteristics, and interactions between them. It can be measured through mental effort, which reflects the amount of controlled processing engaged by the individual. The study found that conventional problem-solving methods, which often involve means-ends analysis, can impose high cognitive load due to extraneous processes not directly relevant to learning. The study also emphasizes the role of practice-problem variability in schema acquisition and transfer. Training with worked examples, especially high-variability ones, was found to be more effective in facilitating schema acquisition and transfer performance. This is because worked examples provide stereotyped solutions and help in rule automation, which are essential for problem-solving transfer. The study used a cognitive-load approach to measure mental effort through rating scales and heart-rate variability spectral analysis. It found that students in the worked conditions perceived less mental effort and achieved better transfer performance than those in conventional conditions. The results suggest that worked examples are more efficient in reducing cognitive load and enhancing transfer performance. The study concludes that effective instructional strategies should focus on reducing cognitive load by directing attention to aspects that facilitate schema acquisition. Worked examples, especially with high variability, are particularly effective in this regard. The findings support the cognitive-load theory and suggest that future research should explore the relationship between cognitive load and learning processes in more depth.