Abstract: An improved cuckoo search (ICS) algorithm is developed to identify temperature dependent thermal conductivity for transient heat conduction problems. Kirchhoff transformation is adopted to transform nonlinear transient heat conduction problems into linear problems. The direct problems are solved by boundary element method. Inversion of thermal conductivity is transformed to estimate unknown coefficients, which is solved by ICS algorithm. ICS algorithm is less sensitive to iterative initialization than conjugate gradient method and ICS algorithm has high efficient convergence compared with cuckoo search (CS) algorithm. For ICS, numerical examples indicate that increase of measured point number causes increase of iteration numbers, whereas increase of nest number decreases iteration numbers. The less the measured noise is, the higher the precision of results is. Iteration numbers decrease at the same time. It shows that ICS algorithm is an accurate and efficient method for identification of thermal conductivity.