The identification of submillimetre phytoplankton is important for monitoring environmental and climate changes, as well as evaluating water for health reasons. Current standard methods for phytoplankton species identification require sample collection and ex situ analysis, an expensive procedure which prevents the rapid identification of phytoplankton outbreaks. To address this, we use a glass-based microchip with a microchannel and waveguide included on a monolithic substrate, and demonstrate its use for identifying phytoplankton species. The microchannel and the specimens inside it are illuminated by laser light from the curved waveguide as algae-laden water is passed through the channel. The intensity distribution of the light collected from the biochip is monitored with an external photodetector. Here, we demonstrate that the characteristics of the photodiode signal from this simple and robust system can provide significant and useful information as to the contents of the channel. Specifically, we show first that the signals are correlated to the size of algae cells. Using a pattern-matching neural network, we demonstrate the successful classification of five algae species with an average 78% positive identification rate. Furthermore, as a proof-of-concept for field-operation, we show that the chip can be used to distinguish between detritus in field-collected water and the toxin-producing cyanobacterium Cyanothece.