On-wing cleaning of engine compressors for commercial aircraft is a required maintenance task which results in greater operating efficiency and lower emission rates. It is typically carried out by injection of water and detergents into the intake of an engine while the engine is being cranked by the starter (i.e. dry-cranked). The dry-ice blasting process, a cleaning system which uses an air-flow and CO 2 dry-ice particles as cleaning agent, has been proposed as an alternative method which is potentially capable of efficient cleaning. In this context, an experimental and numerical investigation of the dry-ice defouling process is presented. A prototype of the new cleaning-system is used to defoul the compressors of a GE CF6-50 engine in a test-rig. The injected dry-ice particles disintegrate into smaller fragments and defoul the airfoils upon collision inside the dry-cranked engine. The study addresses an insight into the dry-ice particle behavior during the process and the numerical assessment of the defouling efficiency. Air-flow measurements, particle-tracking experiments utilizing a high-speed camera (HSC) and airfoil surface-mapping experiments are presented. The particle recordings are made at three positions inside of the engine and these are two-dimensionally postprocessed. The airfoil surface-mapping utilizes photographies and image post-processing to compare the airfoil surfaces before and after cleaning. The airfoils investigated are coated with polyexymethylene, which is used to visualize the defouling effect in large scale engine experiments. The process is simulated in steady state with Ansys CFX and the implementations of the newly developed particle breakup-and erosion models for dry-ice are used in an Euler-Lagrangian formulation. Rotational periodicity is assumed, and the airfoil domains are linked with a mixing-plane approach. Predicted air flow properties of the dry-cranked engine, particle size and particle velocity data, and defouling patterns of the blading are compared using experimental data where possible. Cleaning efficiency is assessed at various instants of time using various parameters of the particle breakup model and of the particle injection formulation. The overall agreement of predicted and experimental data is found to be satisfactory for engineering purposes. The mean deviations encountered range from 9.7 to 22.4%. Possible improvements of the numerical strategy presented are identified and discussed.