In a negative binomial GLM or GLMM what does theta stand for?

When using a GLM or GLMM with a negative binomial distribution in R it is usually necessary to set the value of theta in the model. But what does theta actually stand for? And how do I caluclate it?

I have been able to get around this for now by setting my models to calculate theta for me (for example the glmmADMB package in R can do this) but not all packages can do this. In anycase, I would really like to know what this means so I can understand my results better. Does anyone know?