The pre-clinical validation of cell therapies requires monitoring the biodistribution of transplanted cells in tissues of host organisms. Real-time detection of these cells in the circulatory system and identification of their aggregation state is a crucial piece of information, but necessitates deep penetration and fast imaging with high selectivity, subcellular resolution, and high throughput. In this study, multiphoton-based in-flow detection of human stem cells in whole, unfiltered blood is demonstrated in a microfluidic channel. The approach relies on a multiphoton microscope with diffractive scanning in the direction perpendicular to the flow via a rapidly wavelength-swept laser. Stem cells are labeled with metal oxide harmonic nanoparticles. Thanks to their strong and quasi-instantaneous second harmonic generation (SHG), an imaging rate in excess of 10 000 frames per second is achieved with pixel dwell times of 1 ns, a duration shorter than typical fluorescence lifetimes yet compatible with SHG. Through automated cell identification and segmentation, morphological features of each individual detected event are extracted and cell aggregates are distinguished from isolated cells. This combination of high-speed multiphoton microscopy and high-sensitivity SHG nanoparticle labeling in turbid media promises the detection of rare cells in the bloodstream for assessing novel cell-basedtherapies.