The present study compares the relative effects of two training methods on learner's computer self-efficacy and learning performance in WWW homepage design. A field experiment was conducted by employing two classes of 10th graders. Results indicated that the behavior modeling training method yielded consistently superior performance and higher computer self-efficacy as compared with the instruction-based approach. Subjects with various learning styles performed substantially different in some learning tasks. The significant two-way interaction indicates the critical roles that gender and learning style played in interacting with training method. For learning performance, male students benefitted more from the instruction-based and female students learned better in the behavior modeling condition. Concerning computer self-efficacy, female students gained more from the instruction and male students benefitted more from behavior modeling approaches. For different learning style students, there exists a best-fit training approach. In addition, the best-fit training approach is task dependent. These results suggest that each individual training method has its unique merit to meet designated training objectives for learners with specific traits. Future research directions conclude the paper.