This study investigates whether a tutorial for software training can be enhanced by adding a pedagogical agent, and whether the type of agent matters (i.e., cognitive, motivational, or mixed). The cognitive agent was designed to stimulate students to process their experiences actively. The motivational agent was designed to increase perceived task relevance and self-efficacy beliefs. A mixed agent combined these features. Process and product data were recorded during and after software training of students from the upper grades of vocational education (M ageD 16.2 years). Comparison of scores on performance measures during training revealed a significant advantage of working with the motivational and mixed agents for two important motivational mediators for learning (i.e., strategy systematicity and mood). All students were highly successful during training, improving from an average 30% task completion score on the pretest to a 77% posttest score. On a retention measure 3 weeks later, task completion was still at 66%. Working with the motivational and control agents yielded significantly higher retention scores, whereas working with the motivational and mixed agents led to significantly higher scores on task relevance and self-efficacy beliefs after training. The discussion reflects on the possibilities for improving the internal and external properties of the agents.