Modern enterprises across all spectra are increasingly adopting SOA-based data integration architectures to rapidly respond to transient data business needs. In this chapter, we analyze a new class of enterprise data integration application, called Data Mashup, in which data services are composed on the fly to answer new data business demands. We review the different approaches to data mashup, discuss their limitations and identify the main requirements to data mashup. We propose then a declarative data mashup approach addressing the identified requirements. Finally, we present some research directions that must be followed in order for data mashup technology to mature.