AFM-IR gives access to infrared absorption information at nanoscale lateral resolution. For chemically simple samples (few components, distinct phases, spectroscopically different samples) ratio images and direct evaluation of spectra can often be sufficient to understand chemical composition. When these approaches fail for more complex samples (spectroscopic overlap of components, many components, inter-diffusion) multi- variate methods can be applied to gain deeper insights. Multi-variate methods enable to combine information from multiple spectra or multiple single wavelength images into actual maps of chemical composition – if we are doing it right. “Doing it right” requires an understanding of the signal transduction chain in AFM- IR and the peculiarities of scanning probe microscopy. It also requires to understand optical effects that limit the linear range of AFM-IR [1]. This presentation will discuss the AFM-IR signal transduction chain and which of its parameters can (need to be) controlled to achieve reproducible AFM-IR measurements and what the challenges are in applying chemometric algorithms to AFM-IR datasets. It will also introduce software packages that enable hyperspectral AFM-IR imaging and machine learning.