The thermodynamic properties of steam table are obtained by measurement or approximate calculation under appropriate assumptions. Therefore they are supposed to have basic measurement errors. And thermodynamic properties should be modeled through function approximation for using in numerical analysis. In order to make noised thermodynamic properties corresponding to measurement errors, random numbers are generated, adjusted to appropriate magnitudes and added to original thermodynamic properties. Both neural networks and quadratic spline interpolation method are introduced for function approximation of these modified thermodynamic properties in the saturated water based on pressure and temperature. In analysis spline interpolation method gives much less relative errors than neural networks at both ends of data. Excluding the both ends of data, the relative errors of neural networks is generally within ${pm}0.2%$ and those of spline interpolation method within ${pm}0.5$~1.5%. This means that the neural networks give smaller relative errors compared with quadratic spline interpolation method within range of use. From this fact it was confirmed that the neural networks trace the original values better than the quadratic interpolation method and neural networks are more appropriate method in modelling the saturated steam table.