Upcoming regulations for vehicle exhaust emission demand substantial reduction of particle emission in diesel exhaust. To achieve these emission levels, the car manufacturing industry is developing new combustion concepts and exhaust after-treatment techniques such as the use of catalysts and particle filters. Many of the state-of-the-art analytical instruments do not meet the required detection limits, in combination with a high temporal resolution necessary for engine optimization. This paper reports a new detection system and the first results of its application to on-line diesel exhaust soot measurements on a engine test bench (MAN diesel engine facility Nürnberg, Germany). The instrument is based on differential photoacoustic (PA) spectroscopy of black carbon aerosol. It contains two identical PA cells, one for the measurement of the aerosol particles and one which analyses the particle-free gas. Thus, a potential cross-sensitivity to gaseous absorbers in the exhaust gas can be excluded. The PA cells were characterized in a laboratory set-up, with water vapor as reference gas and artificial soot generated by a spark discharge generator. The detection limit was found to be 2 microg m(-3) BC (for diesel soot) with a sampling rate of 3 Hz. The temporal response of the system was found to be in the order of 1 s. After full characterization of the cells, the system was transferred into a mobile 19"-rack. Characterization of the mobile sensor system under real-world conditions was performed during several measurement campaigns at an engine test bench for heavy-duty diesel engines. Results for the limit of detection, the time resolution, accuracy, repeatability, and robustness of the sensor system are very promising with regards to a routine application of the system in engine development.