Purpose: The purpose of this article is to provide trial-by-trial practice performance data in relation to learning (outcome probe data) as collected from 18 treatment sessions provided to children with severe speech sound disorders. The data illustrate the practice-learning paradox: Specific, perfect practice performance is not required for speech production learning. Method: We detailed how nine student speech-language pathologists (SSLPs) implemented and modified the motor learning practice conditions to reach a proposed challenge point during speech practice. Eleven participants diagnosed with a severe speech sound disorder received high-intensity speech therapy 3 times per week for 6 weeks. SSLPs implemented treatment procedures with the goal of achieving at least 100 practice trials while manipulating practice parameters to maintain practice at the challenge point. Specifically, child performance was monitored for accuracy in five-trial increments, and practice parameters were changed to increase functional task difficulty when the child's performance was high (four or five correct responses) or to decrease functional task difficulty when the child's performance was low (fewer than four correct responses). The practice stimulus, type and amount of feedback, structure of practice, or level of support might be changed to ensure practice at the challenge point. Results: On average, the children achieved 102 practice trials per session at a level of 58% correct responses. Fast achievement of connected speech with the lowest amount of support was associated with high scores on generalization probes. Even with high levels of error during practice, the children improved percent consonants correct with maintenance of learning 3 months posttreatment. Conclusion: The results of this study show that it may not be necessary to overpractice or maintain a high degree of performance accuracy during treatment sessions to achieve transfer and retention of speech production learning.