The purpose of this study is to investigate effects of use of voice in interface agents included in educational software on students' agent preferences and learning levels. In this context, it was investigated to see whether there are any relations between student gender and the preferred agent gender; students' preference among agents with voices in different frequencies; and effects of the voicing technique on students' achievements. The post-test experimental model was used in the study and the study group consisted of 112 primary education students. Data collection tools of the study included pre-knowledge test; four educational software packages, in which agents having different voices and genders were used, and an achievement test. Multi-factor ANOVA and ANCOVA tests were used in analysis of data. At the end of the study, it was seen that the frequency used in voicing the pedagogical interface agents of the software did not have any significant effects on students' learning levels, and it was observed that students tend to choose the agents of their own gender, and that they prefer high-pitched voice for male agents and high-pitched voice for female agents.