Assessing phytoplankton diversity is of primary importance for both basic and applied ecological studies. Following the advances in molecular methods, phytoplankton studies are switching from using classical microscopy to high throughput sequencing approaches. However, methodological comparisons of these approaches have rarely been reported. In this study, we compared the two methods, using a unique dataset of multiple water samples taken from a natural freshwater environment. Environmental DNA was extracted from 300 water samples collected weekly during 20 years, followed by high throughput sequencing of amplicons from the 16S and 18S rRNA hypervariable regions. For each water sample, phytoplankton diversity was also estimated using light microscopy. Our study indicates that species compositions detected by light microscopy and 454 high throughput sequencing do not always match. High throughput sequencing detected more rare species and picoplankton than light microscopy, and thus gave a better assessment of phytoplankton diversity. However, when compared to light microscopy, high throughput sequencing of 16S and 18S rRNA amplicons did not adequately identify phytoplankton at the species level. In summary, our study recommends a combined strategy using both morphological and molecular techniques.