In present paper, the inverse problem of cavity identification, i.e. determining the location and geometric shape of a cavity inside a two dimensional elastic body using temperature measurements obtained by performing a simple heat conduction test is investigated. The Boundary Element Method (BEM) coupled with Genetic Algorithm (GA) and Conjugate Gradient Method (CGM) is used to solve this inverse problem. A fitness function which is the squared differences between measured temperatures and calculated ones at the same locations on the exterior boundary of a sample containing a cavity is minimized using CGM. Considering that inverse problems are generally ill-posed and local optimization methods require a good initial guess, a mete heuristic model based on GA for determining a good initial guess of the cavity is presented in this study. Then this initial guess is used by CGM to achieve accurate location and geometric shape of cavity. The major cases application of this research is in the casting industry.