Early and accurate detection of parasitic infections is important for diagnosis, particularly in a resource-limited setting. This study presents an automatic classification approach using convolutional neural networks for both a binary and multi-class classification problem, to differentiate presence of infection and explicitly classify parasitic species. On a dataset consisting of 25,041 microscopic images of nine parasitic species (Cryptosporidium sp., Entamoeba histolytica, Giardia lamblia, Leishmania sp., Toxoplasma gondii, Trichomonad, Enterobius vermicularis, Hymenolepis nana and Isospora sp.) for which species included 30 to over 10,000 images we also evaluated the framework. We solved class imbalance by using the advanced data augmentation processes or transfer learning with the pre-trained models (MobileNet, DenseNet, Xception, etc.) and got an accuracy of 98%. The clinical relevance of the approach was demonstrated by high recall rates for challenging species such as Toxoplasma gondii and Leishmania sp. Strong performance notwithstanding, error analysis suggested some improvement is needed for underrepresented species like Isospora sp. This work develops a scalable and flexible system for parasitological diagnostics suitable for a wide array of clinical needs, particularly in low-resource settings.