SUMMARY: Identification of duplicate templates is a common preprocessing step in bulk sequence analysis; for large libraries, this can be resource intensive. Here, we present streammd: a fast, memory-efficient, single-pass duplicate marker operating on the principle of a Bloom filter. streammd closely reproduces outputs from Picard MarkDuplicates while being substantially faster, and requires much less memory than SAMBLASTER. AVAILABILITY AND IMPLEMENTATION: streammd is a C++ program available from GitHub https://github.com/delocalizer/streammd under the MIT license.