The performance optimization of modern Spark Ignition engines is limited by knock occurrence: heavily downsized engines often are forced to work in the Knock-Limited Spark Advance (KLSA) range. Knock control systems monitor the combustion process, allowing to achieve a proper compromise between performance and reliability. Combustion monitoring is usually carried out by means of accelerometers or ion sensing systems, but recently the use of cylinder pressure sensors is also becoming established, especially for motorsport applications. The cylinder pressure signal is often available in a calibration environment, where SA feedback control is used to avoid damages to the engine during automatic calibration. Since the knock phenomenon is influenced by several factors, the determination of KLSA should be carried out based on the observation of the engine actual behavior: to achieve this, a general model could be runtime-adapted to the present running conditions, based on available feedback data. Several knock models are available in the literature: most of those proposed for real-time applications are single zone or two-zone models, grounded on more complex CFD simulations. The present paper proposes a simplified approach, based on a statistical analysis of combustion data: typical indicating analysis outputs available cycle-by-cycle and cylinder-by-cylinder are processed, in order to represent the knock intensity sensitivity to the combustion phasing. The combustion phasing is then related to SA, in order to evaluate the KLSA, that is determined by setting a threshold on the probability of reaching given knock intensity limits. The approach has been applied to indicating data referring to several engines, showing a good prediction capability.