Quantitative identification algorithm of defect or defect boundary is the key theoretical basis for the development of quantitative nondestructive testing and evaluation. For the geometry identification of inner pipe boundary (i.e. geometry-varied boundary due to defects), most methods developed up to now are not precise for the identification of irregular-shaped inner boundary. In this paper the steady-state identification problem was solved, with different inner boundary geometries and different test piece shapes, based on finite element method and conjugate gradient method by associating ANSYS software with MATLAB. A series of effecting factors, such as the initial boundary geometry guess, temperature measurement error and number of measurement points at the inspection boundary, number of discrete points of the boundary to be identified, and the thermal conductivity of the test piece, were studied systematically on the effect on the identification result. Numerical experiments certified the effectiveness of the method and proved the method to be very precise and time saving.