Thermal storage units, particularly stratified water tanks, are crucial for optimizing energy systems and smart buildings. However, accurately modelling these systems poses challenges, for example within Model Predictive Control (MPC) frameworks. This study explores the trade-offs between complexity and accuracy in simplified grey-box Resistance-Capacitance (RC) thermal models of stratified water tanks for HVAC control purposes. A comparative analysis of one-node and two-node RC models was conducted, validating them against experimental data from three tank configurations: a tank with coiled heat exchanger, a fully mixed tank, and a tank with electrical resistances. These were tested under various operating conditions, i.e. standard domestic hot water (DHW) tapping profiles, in a laboratory-controlled environment. Results show that optimized one-node models provide reasonable accuracy while maintaining computational simplicity. Two-node models, despite increased complexity, did not consistently outperform optimized one-node models, as fixed-volume nodes struggled to capture temperature stratification accurately. Model performance varied with tank configuration and operating conditions. Tanks with higher stratification levels showed more pronounced differences between one-node and two-node model performances. In conclusion, while simplified grey-box models can effectively represent stratified water tank dynamics, model complexity selection should consider factors like calibration data, desired accuracy, computational resources, and stratification levels. These findings provide valuable guidance for model selection and MPC strategies in HVAC systems with thermal storage.