One-on-one tutoring is a form of instruction that requires interaction between a tutor and a tutee. The effectiveness of tutoring is examined from the perspectives of the tutor's actions, the tutee's actions, and successive interactions. A tutor's actions that may not lead to successive interactions consist of asking an initiating question, providing feedback, and asking a comprehension-gauging question. It is suggested that these types of actions can lead to the learning of an ideal template of solution procedures for solving problems, but may not lead to deep understanding. A tutee's actions are postulated as self-explaining, in response to either tutors' questioning, tutors' prompting, or tutors' scaffolding. Finally, an interaction is defined as a sequence of tutor actions, such as scaffolding, which elicits a successive series of exchanges between the tutor and tutee, so that they collaboratively construct a response. An exercise in a detailed protocol analysis of a case study of a student being tutored in solving a mechanics problem is presented. It shows what misconceptions the student exhibited, whether these misconceptions were removed, and what actions triggered the learning. The results of this case study support the suggestion that tutor actions that prompt for co-construction (which includes self-explanation) may be the most beneficial in producing deep learning, in the sense of removing misconceptions.