Background: Teaching programming and computational thinking is becoming a major issue in many education systems. Numerous approaches are possible, but very few studies compare these different ways of implementing programming and computational thinking learning. Objectives: We compared three ways of teaching programming and computational thinking to grade 5 students: unplugged activities, block-based programming software and educational robotics. Method: This paper re-exploits previously published data on the comparison between unplugged and plugged-in (using the block-based programming software Scratch) learning, by adding a new experimental condition centred on the use of Scratch associated with an educational robot, and incorporates new mixed model statistical analyses. Based on these results, we conducted a second experiment to explore the cognitive load associated with a learning situation on Scratch, with or without an educational robot. Results: Our first experiment revealed that novice grade 5 students using Scratch without a robot learned better programming concepts and skills, compared to students involved in unplugged activities or using Scratch with a robot. Experiment 2 showed that the detrimental effect of using a robot is probably due to an increased extraneous cognitive load in the Scratch with a robot condition. However, robots had a large positive effect on students' motivation. Gender differences were also observed, as girls showed slightly better learning performance but were less motivated than boys when using Scratch without a robot. Conclusions: Overall, these results provide key elements for understanding the advantages and disadvantages of using robots to sustain computational thinking learning in children.