Principal component analysis was conducted on five major climatic variables-dry-bulb temperature, wet-bulb temperature, global solar radiation, clearness index and wind speed. Twenty-eight year (1996-2000) long-term measured weather data were considered. A two-component solution was obtained, which could explain 80% of the variance in the original weather data. Monthly electricity consumption data recorded during a 5-year period (1979-2006) were gathered from 20 fully air-conditioned office buildings with centralised HVAC systems in subtropical Hong Kong. Electricity use per unit gross floor area ranged from 163 to 389 kWh/m(2). These consumption data were correlated with the corresponding principal components using linear multiple regression techniques. The coefficient of determination (R-2) varied from 0.76 to 0.95 indicating reasonably strong correlation. It was found that the regression models developed could give a reasonably good indication (mostly within 3%) of the annual electricity use, but the monthly estimates might differ from the actual consumption by up to 9%. Attempt was also made to develop a general regression model for the 20 buildings, which had an R-2 of 0.84 with a maximum mean-biased error of 18.6% and a maximum root-mean-square error of 21.4%. (C) 2007 Elsevier B.V. All rights reserved.