This research aimed to investigate the effect of nanorice husk ash (NRHA) prepared using different thermal treatment methods on ultra-high-performance concrete (UHPC) behaviour. NRHA was prepared by two methods: (1) burning for 3 h at 300, 500, 700 and 900 °C and (2) burning for different durations (9, 7, 5 and 3 h) at 300, 500, 700 and 900 °C. NRHA was added to UHPC to make 25 mixtures with three dosages (1%, 3% and 5%). Density, compressive strength, tensile strength, flexure strength and ultrasonic pulse velocity tests were performed at the experimental level. Moreover, full microstructure analysis, including X-ray diffractometry, Brunauer-Emmett-Teller surface area analysis, thermogravimetric analysis, scanning electron microscopy and energy-dispersive X-ray spectroscopy, was performed. The best performances in in the first method (constant duration, different temperatures) were obtained by 1% NRHA burned at 900 °C with 12.5% compressive strength and 1% NRHA burned at 700 °C with increased ratio (10%). Moreover, the best performance in the second method (different burning durations and temperatures) was obtained by 3% NRHA with a ratio of 22.5% at 700 °C for 5 h. Burning rice husk ash improved the compressive strength. It also remarkably improved the splitting tensile strength and flexure strength by 32% and 47%, respectively, at 3% NRHA treated at 700 °C for 3 h. The microstructural analysis showed the efficient role of NRHA in the compactness of concrete sections. It improved the formation of new calcium silicate hydrate gel; decreased the cracks, voids, CaCO3 and Ca(OH)2; and increased the Ca/Si composition. The obtained experimental results were used to build an artificial neural network (ANN) to predict UHPC properties. The ANN model was used as a validation tool to determine the correlation between results. Results showed a remarkable improvement in the mechanical properties of UHPC incorporating NRHA for all mixtures. The ANN model indicated a reliable correlation between input and output variables. The R2 values for the training, validation and testing steps were all 0.99.