We demonstrate a label-free cell analyzer with high throughput and single-cell resolution based on optofluidic time-stretch microscopy. It is capable of acquiring images of cells at a throughput beyond 10,000 cells/s with a spatial resolution of 780 nm in a label-free manner, allowing for statistical analysis of the cells' phenotypes from the images via machine learning methods. Moreover, using compressed sensing, the throughput can be further improved by a factor of 50. Thus, our scheme holds great potential for cell analysis in various scientific and industrial fields, including bio-fuel production and disease evaluation.