You are unlikely to create the final model in your first attempt. Use the power of multivariate analysis to select interesting samples based on the instrumental measurements before you put them through the lab. This will save you a lot of time and money by identifying those samples which are the ‘needles in the haystack’ i.e. the important ones.
This is a key step. Use various plotting tools to isolate bad situations before you analyze the data, otherwise your analysis will be poor. Depending on the type of data use line- 2D-, 3D- or matrix-plots. If you have multiple responses be sure to visualize the correlation structure among them. On the other hand, by looking at raw data, the human eye can not decide if there is enough information to model your response variables.