Despite considerable advantages such as reliable hardware, no disturbance in production, fast analysis and presentation of results and very low operational cost, the use of percussive drill monitoring to retrieve geological and geomechanical information concerns the drilled rock has not become a standard investigation tool in the underground industry. One major obstacle is the analysis of data. The monitored ``raw'' data can be aected by a signi®cant in¯uence from the operator, who often adjusts the drill settings in order to achieve the best drilling result. Furthermore, the advanced control system on a modern drill rig adjusts drill parameters independently to avoid drilling problems and damage to the drill string and machine. In order to use percussive drill monitoring in industrial applications, the performed analysis must be able to handle variations among monitored drill parameters, but still separate rock dependent variation from other in¯uences on the monitored drilling data. Once the external in¯uences on the drilling data have been normalised only leaving the rock dependant variation, rock properties can be predicted using theoretical or heuristic relations, or calibrated with observed rock conditions using statistical analysis. The method of analysis suggested in this paper is based on a step-wise normalisation of raw drilling data, where hole length dependent variation initially is removed, followed by a normalisation of the thrust dependent variation, and ®nally, by removing the in-¯uence of penetration rate on torque pressure. The analysis shows that major classi®cation errors can be made if raw data are used instead of normalised data. The technique is applied on modern drill monitoring data from three dierent drilling sites, in crystalline rock masses in Sweden. The drill data selected were obtained during normal drilling conditions, where no special steps had been taken to promote the analysis or the interpretation. Examples from each site show good potential to predict ore boundaries and fracturing based on normalised drill parameters from percussive drilling. # 1998 Elsevier Science Ltd.