Accurate recognition and classification of peripheral blood cells are crucial for hematological diagnostics. While previous studies applying convolutional neural networks to blood images mainly focused on classification, this study evaluated the potential of object detection models. We manually annotated a large dataset of blood smear images and employed YOLOv5s, SSD300, and FRCNN to detect eight types of blood cells. The results showed that YOLOv5s outperformed all models with the highest mAP@0.5 of 98.9%. Both YOLOv5s and FRCNN achieved the highest sensitivity of 97.0%. However, as the object detection approach accounts for the actual imbalanced distribution and sizes of cells, particularly platelets, model performance on this cell type was generally poorer. This evaluation highlights the effectiveness of object detection models, especially YOLOv5s, for blood cell recognition while emphasizing the need for further improvements in CNN-based approaches to address the complex characteristics of blood cell images.