In this paper, mixed cross-sectional data and time series data are employed to explore the relationship between public health expenditure and income as well as other factors which have nothing to do with income in China from 1991 to 2010. In order to study the stationarity of these variables, this paper does not use the critical value of the asymptotic normal distribution, but conducts panel LMunit root tests to reflect the structural changes using finite sample critical values obtained by the Bootstrap method. It is an important conclusion that the government budget deficit can exert long-term significant impact on public health expenditures. So, this paper provides supporting evidence on significant difference in the fiscal policy of public health as well as the health service coverage between rich and poor areas.