A general autoregressive conditional heteroskedasticity (GARCH)-typed variance-gamma (VG) model, the VG-GARCH model, developed by Kao, Wu, and Lee (2011) is adopted for the analysis of high-frequency intraday price data in a price-duration setting. The advantages of the proposed VG-GARCH model are in three respects: (1) The price process is based on an Ito's semi-martingale, namely, a VG process, which is a desirable feature in finance theory; (2) The VG process has its physical-economical implication in that it can be expressed as the market liquidity measure VNET defined by Engle and Lange (2001); (3) By allowing time-varying scale parameter that obeys a nonlinear GARCH (NGARCH) process, the feedback and leverage effects of price changes on elapsed time, and the resulting temporal dependency and clustering in elapsed times and price changes are accounted for. An empirical study is implemented by comparing the proposed model to the EACD-VAR model by Dufour and Engle (2000) of using one-minute S&P 500 index from January 2, 2001 to May 31, 2001. The performance is demonstrated, specifically through a comparison of the proposed model and the competing model.