To be able to validate models based on signals logged from processes, it is necessary to have a continuous data set. Since missing data is a common problem when logging process variables, it can be almost impossible to find a continuous data set. In general it might be necessary to replace missing data also when estimating models for the signals and this paper compares different approaches for missing data estimation. The result shows that on the available data set there is no significant difference in the performance and for the model validation problem it is probably enough to use the simplest method. However, for other applications and on other data sets, there are probably important differences.