The power grid has seen record demand for frequency regulation capacity in recent years due to the increased employment of renewable energy resources worldwide. Building thermal loads are flexible and thus, can be used as regulation reserves with proper control strategies. Previous studies have shown that building HVAC equipment can provide high-quality power grid regulation service with PJM performance scores of up to 0.98 and buildings' participation in the regulation market could bring significant economic benefits for building owners. However, the power flexibility in buildings is not persistent and the available HVAC regulation capacity has significant hour-by-hour variation due to building load and other operating constraints. This paper presents a regulation capacity reset strategy for HVAC regulation control that identifies the available regulation capacity and baseline power on the fly with real-time load and operation data. The strategy relies on a steady-state HVAC performance model derived from manufacturer performance data and implements a pseudo-optimization that seeks the maximum regulation capacity while respecting all operating constraints. The proposed strategy was implemented on a variable-speed rooftop unit (RTU) and validated with laboratory tests in psychrometric chambers. The test results show that the proposed reset strategy is effective in providing consistent high-quality regulation service with negligible impact on the indoor temperature control; the zone temperature deviation from the setpoint was within 0.3 degrees C for all the performed tests. The reset strategy was also simulated with a prototypical building diurnal load model to quantify the integrated regulation capacity for a typical summer day. Simulation results indicate the integrated HVAC regulation capacity throughout a summer day equals approximately 1/4 of the daily electrical energy use; and the estimated daily regulation credit can offset up to 26% of the daily HVAC electricity cost based on PJM historical prices. (C) 2019 Elsevier B.V. All rights reserved.