NUPACK is a RNA secondary structure prediction package, used widely by many biologists. For its thermodynamic model, it uses the Turner 2004 parameters. One of the limitations of the Turner 2004 parameters is that it does not take into account much of the experimental data that is now available. Constraint Generation(CG) software, developed at the Beta lab, has been successful in training the Turner parameters for Hotknots(structure prediction software developed at Beta lab), and has reported better accuracy. My talk will focus on how we have used CG to train parameters for NUPACK, and the challenges associated with this.