We cite data for the same reasons we cite anything else:  to give credit to those who made the data available and to help our readers find the data we used.  A good citation answers 4 basic questions:

• Who collected, produced, or provided this resource? 
  • Authors,  researchers, data collectors, and/or organizations that sponsored the research.
• What is this resource? 
  • A title, the data's version or edition, the format or resource type.
• When was this resource collected, created, or made available? 
  • A dataset may have a date-of-collection as well as a publication date.
• Where can someone find this resource?
  • The website, organization, and/or publisher that provides access to the data. Whenever possible, this should include an identifier like a DOI or a URL for the data's website. Make sure your citation includes enough information for a reader to find the data easily!

The details you need to answer these questions will be different, depending on the data you are using.  It's not an exact science, but as a general rule, a good citation will answer these questions accurately and thoroughly.  It's better to have too much information than too little!

The APA Style Guide provides a recommended citation format for databases, with examples, but other style guides, including MLA and Chicago, don't--so you'll have to create your own.

For citation styles that do not have a specific dataset format, you can base your citation on the closest equivalent formats. For example, if the dataset is online, use a format for online items. If it was created by multiple "authors" or editors or researchers, use a format for edited works or items with more than one author.  Or, just base your citation off of the general reference format for your style guide.

If a grant application, scholarly journal, or instructor has strict requirements for citations, they'll usually make this clear, and may provide examples.  If they don't offer an example for citing the kind of dataset you are using, don't be afraid to ask!