MOTIVATION: The Sequence Read Archive (SRA) contains raw data from many different types of sequence projects. As of 2017, the SRA contained approximately ten petabases of DNA sequence (10 16 bp). Annotations of the data are provided by the submitter, and mining the data in the SRA is complicated by both the amount of data and the detail within those annotations. Here, we introduce PARTIE, a partition engine optimized to differentiate sequence read data into metagenomic (random) and amplicon (targeted) sequence data sets. RESULTS: PARTIE subsamples reads from the sequencing file and calculates four different statistics: k -mer frequency, 16S abundance, prokaryotic- and viral-read abundance. These metrics are used to create a RandomForest decision tree to classify the sequencing data, and PARTIE provides mechanisms for both supervised and unsupervised classification. We demonstrate the accuracy of PARTIE for classifying SRA data, discuss the probable error rates in the SRA annotations and introduce a resource assessing SRA data. AVAILABILITY AND IMPLEMENTATION: PARTIE and reclassified metagenome SRA entries are available from https://github.com/linsalrob/partie. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.