Below is a selection of recently published results from reasonably recognized firms - no fly-by-night CEPS/Survey Fasts here.
Of course, each of these polls uses a different methodology, sampling procedure and likely-voter estimation. They cover different geographical areas and treat the "won't respond" group differently. This is not a time-series. Results aren't comparable to one another, so there's no use looking for trends; it's the absolute numbers that are interesting.
One thing I find interesting is that there's a much wider spread between Rosales's high and low numbers (from 24 to 48) than between Chavez's (45-58). But then, that Rosales 48% comes from Keller, which was not really a voting intentions survey - so if you overlook that one, Rosales's best showings are in the low 40s - PSB and C21.