XML provides flexible, extensible data models and type systems for structured data, and has found wide-acceptance in many domains. XML processing can be slow, however, especially for scientific data, thus leading to the conventional wisdom that XML is not appropriate for such data. Instead, data is stored in specialized binary formats, and is transmitted via work-arounds such as attachments and base64 encoding. Though these work-arounds can be useful, they nonetheless relegate scientific data to second-class status within the web services framework; and they generally require yet another API, data model, and type system. An alternative solution is to use more efficient encodings of XML, often known as "binary XML". Using XML uniformly throughout an application simplifies and unifies design and development. In this paper we present a binary XML format and implementation for scientific data called Binary XML for Scientific Applications (BXSA). We show that performance is comparable to that of commonly used scientific data formats such as netCDF. These results challenge the prevailing practice of handling control and data separately in scientific applications, with web services for control and specialized binary formats for data.