They claim to be legit, even to a +/- 1.95% MOE. It appears the sample is self-selected, which makes that seem unlikely, but they claim to make up for it with their weighting methodology. I selected myself as part of the sample and completed a survey. The number of grammatical errors in the questions leads me to take the survey less seriously than I otherwise would.
The MOE only refers to the information derived from the sample. So it is a near certainty that the sample is very pro Romney but this says nothing about the relevance of the sample with regard to the actual population.