There are numerous datasets available, but all have some limitations. The most famous ones are (still) DARPA 98/99 and KDD99, but they have several shortcomings and have been criticized a lot, e.g., by Mahoney and Chan (e.g., see http://link.springer.com/chapter/10.1007/978-3-540-45248-5_13). Even so, they are still used today, but results and evaluations done by these datasets are quite worthless. Some improvements had been done by NSL-KDD as already mentioned; Qian et al. presented another redesign of the DARPA set (see http://link.springer.com/chapter/10.1007%2F11766155_32).
Real-world data can be found at the MAWI Working Group Traffic Archive and the WIDE project (see http://mawi.wide.ad.jp/mawi/). While real-world (recorded) data is a good challenge for a realistic comparison of IDSs, it often suffers from a missing ground truth. Further packet-, flow- and http- traces can be found at MOME (Cluster of European Projects aimed at Monitoring and Measurement, http://www.ist-mome.org/database/MeasurementData/index8db7.html) or Consortium Internet 2 (you may have to mail them because the public link seems to be removed?), CAIDA (http://www.caida.org/data/overview), the Waikato Internet Traffic Storage
(http://wand.net.nz/wits/), RIPE (https://labs.ripe.net/datarepository/data-sets), the Internet Traffic Archive (ita.ee.lbl.gov/html/traces.html), the UMassTraceRepository (http://traces.cs.umass.edu/index.php/Network/Network) or PREDICT (http://www.predict.org/Default.aspx?tabid=175).
For flow-based systems, labeled datasets can be found at http://www.simpleweb.org/wiki/Labeled_Dataset_for_Intrusion_Detection
As already mentioned, all available datasets have different limitations; e.g., see https://tnc2014.terena.org/core/poster/13 for an overview of some pros and cons.
Why not make your Own dataset? I think most people would agree when I say " Each network is different, and the type of attacks and attack durations will also be different depending on the services you have. So a dataset that is good for one scenario is not going to good for another and so are the attributes. Which is why we created this tool which allows you to make you own intrusion dataset and you get to choose your attributes!
Paper: Conference Paper INSecS-DCS: A Highly Customizable Network Intrusion Dataset ...