Recent theoretical and experimental progress towards the achievement of an autonomous control system for spacecraft nuclear reactors is reported. A controller having a multitiered structure with the capability for both supervisory and predictive control as well as automated reasoning is proposed. Such a system would establish bounds for permitted operation (supervisory control), facilitate the inference of system response (predictive control), and provide a means for assessing its own performance (automated reasoning). Recent developments concerning specific features of this controller are reported including the use of energy constraints for supervisory control, the performance of automated reactor startups using the MIT-SNL Period-Generated Minimum Time Laws, and considerations relevant to the use of pattern recognition, expert systems, and causal analysis for the on-line assessment of controller performance. Also, an overview is given of the MIT program on reactor control.