This paper estimates linear structural models using LISREL and employs MIMIC models to find out factors determining child health in Pakistan. A distinction has been made in permanent and transitory health states that lend support to Grossman's (1972) stock and flow concepts of health. The paper addresses the issue of health unobservability and finds out that latent variables using MIMIC models best represent underlying child health states. To overcome problems of poor income data, factor analysis is applied to extract measures of housing and durables as indicators of socio-economic well-being of children in Pakistan. The results of the study show that child health states, both permanent and transitory, are affected significantly by factors such as parental education, socio-economic conditions, and health care variables.