Computer training is vital for improving user interaction with a system. Prior research on computer training indicates that behavior modeling is highly effective for computer skill acquisition and can be further improved by incorporating symbolic coding processes. Building upon dual coding theory, our study proposes two new coding strategies, imagery coding and procedural coding, and examines the effects of those alternative coding strategies on computer learning outcomes relative to textual coding in a modeling-based training context. A training program on a popular computer graphic application was offered in a university setting and attended by 190 trainees. The study results show that modeling-based training can be significantly improved by incorporating the proposed new coding approaches but the specific coding strategy should be carefully chosen depending on the target learning outcome.