Pollen monitoring helps predict the risk of pollen-induced allergies. Traditionally, this monitoring is performed based on the biological material obtained from volumetric Hirst’s traps. A palynological specialist analyzes the obtained microbiological specimen under the microscope, and recognizes and counts pollen grains of various taxa. This is a tedious task, and automatic detection and counting of pollen grains in digital microscopic images can support specialists in their work. YOLOv5 and Faster R-CNN are the state-of-art deep neural networks used for object detection in many fields of computer vision. In the presented research, these detectors were applied to analyze specimen with pollen grains of four taxa, typical of early spring in Central and Eastern Europe. The obtained results enabled the selection of the detector that should be the first choice in pollen grains recognition tasks. Statistical analysis of differences in the distribution of the recognition quality measures also supports the conclusions.