The number of signal rays passing through a research area are often limited due to the geometric distribution of satellite constellations and receivers. To solve this issue, an approach is proposed that imposes virtual slant signals on the tomographic model. This approach increases the number of signal rays passing through a research area and. improves accuracy. An experiment tested the feasibility and accuracy of proposed approach by comparing with the results using Global Positioning System and meteorological data for 31 days in May, 2013, provided by Hong Kong Satellite Positioning Reference Station Network and those derived from radiosonde data. The Root Mean Square water vapor profile and error statistics for proposed method are all superior to that of traditional method. The experiment shows that the proposed approach can improve the accuracy of tomographic results.