Contents of Ca and Mg in the slag were analyzed quantitatively with laser-induced breakdown spectroscopy (LIBS). Due to the complexity of the slag composition a regression relationship established often fails to get the desired result, this results in that the problem of multiple variable regression analysis must be considered. In order to analyze the contents of Ca and Mg in the slag, the Mg, Ca, Fe, Si, Al atomic line intensity and Mg, Ca ion line intensity were used as the input vectors. However, when the absolute line intensity including strong spectral lines and weaker spectral lines are put together as the input vector, the influence of the former will cover up the latter. Spectral intensity needs treated first to place all the values of the line intensities in a similar range. The slag composition analysis were then performed and compared using three calibration methods like nonlinear multi-function, BP neural network and radial basis function (RBF) network. In addition, the advantage of the RBF neural network calibration relative to traditional non-linear calibration method was particularly emphasized.