A really nice post was done today by DrugMonkey. I think that he issues that DrugMonkey raises are even more important in epidemiology than in bench science. In bench science you have the possibility of replication in a strict sense (and it is, in fact, a governing principle of bench science). In Epidemiology, the population changes between studies and it is difficult to compare between populations. As Heraclitus said "You can never step into the same river twice"; in Epidemiology you never have the same study population twice.
Perversely, this fact actually makes the methods more important as failure to replicate could also indicate unique features of a population. This makes it critical to be able to separate methodological issues from population issues, insofar as this is possible.
Given how badly medical papers seem to document decisions (part of it being a style issue for the field), there is a lot we could do to improve on matters.