The purpose of this pilot study with a within-subject design was to gain a deeper understanding about the promise and restrictions of a virtual tutoring system designed to teach science to first grade students in Finland. Participants were 61 students who received six tutoring science sessions of approximately 20 min each. Sessions consisted of a sequence of narrated multimedia science presentations during which a virtual tutor explained science phenomena displayed in pictures. Narrated science explanations were followed by one or more multiple choice questions with immediate feedback about students' choices and a possible second attempt, during which students reached 97% accuracy. A pretest and posttest was administered to assess students' ability to reason about the science and to transfer knowledge to new contexts. Results indicated significantly greater improvement in the understanding of the science concepts taught during the tutoring sessions, relative to the concepts that were not taught. Results from the surveys administered to teachers and students indicated that the program was well received. Detailed analysis of student error responses provided a deeper understanding about the complex interplay between students' prior knowledge, the way topics were taught in the multimedia lessons, and the way learning was assessed. Findings from the quantitative and qualitative analyses are discussed in the context of designing high quality lessons delivered through a virtual tutoring system.