A ground-based Counterflow Virtual Impactor (CVI) was optimized to achieve nearly complete in situ segregation of cloud droplets and ice crystals (with subsequent evaporation, releasing dissolved gaseous and non-volatile material) from their surrounding carrier gas and interstitial aerosol particles. With a one-dimensional numerical model, the CVI cut size D-50 was reduced to 4 mu m from 7 mu m in an earlier design (Anderson et ai., 1993). This could be achieved by a velocity increase to 225 m s(-1) inside the wind tunnel forming part of the ground-based CVI, and by minimizing ail dimensions contributing to the stagnation length L-stag (distance from the wind intersection plane tunnel:CVI to the stagnation plane inside the CVI that cloud elements have to reach to be sampled). CVI and high-speed wind tunnel were designed and constructed according to the modeling results. Subsequent calibrations verified the: calculated lower cut sizes D-50 and quantified the slope of the collection efficiency curve in terms of cut sharpness S-cut. With the new CVI lower cut sizes between 4 and 6 mu m can be achieved. A cloud chamber experiment was performed with CVI measurements supplemented by a Forward Scattering Spectrometer Probe (FSSP). It could be demonstrated that significant drop break up is caused by wind tunnel velocities well beyond 150 m s(-1). For a reduced wind tunnel velocity of 150 m s(-1) a reasonable cut size of at least 5 mu m could be maintained, while avoiding break-up. The demonstration of break-up should have consequences For any cloud sampling technique featuring high relative velocities of cloudy air past the inlet. In particular. in-cloud retrieval of cloud nuclei concentrations on high-speed airborne platforms could be affected to a significant extent. (C) 2000 Elsevier Science Ltd. All rights reserved.