The identification of jets originated from b quarks is crucial for a broad range of physics analyses. Various b tagging algorithms exist at CMS and are further developed with the use of the machine learning techniques. Constant monitoring of the basic quantities provided to the high-level taggers is fundamental to ensure a good tagging performance and to spot potential issues in the data taking. We present a comparison between the proton-proton collision data collected by the CMS detector in 2016 and simulation. The comparison is between the input variables used by the heavy flavour tagging algorithms and the taggers distributions in several event topologies.