We propose a new method for thermal rendering in telepresence systems which allows an operator to feel most transparently the thermal behavior of the remote object. It is based on learning and thermal heat flux generation. Two databases are constructed from real measurements recorded during direct contact between operator's finger and different materials. One database is used in order to identify the material in the slave side, the other database is used to generate desired heat flux for the thermal display loop. The identification bloc is based on Principal Component Analysis and Neural Network. Experimental results validating the proposed method are discussed.