Helpful reviews have two things in common, according to Leonie Mueck, Alicia Newton, and Sebastien Thuault, senior editors at Nature, Nature Geoscience, and Nature Neuroscience, respectively.
While they note at the Schemes and Memes blog that "[i]t's difficult to generalize what makes reviews particularly helpful, since each manuscript has different strengths and weaknesses," they say that one key component of a good peer review is being able to back up any claims. That is, that if the review says the methodology is flawed, it includes details as to how. The second is component keeping a constructive tone, they add.
In addition, the trio of editors notes that peer reviews have two fundamental charges: to ensure the paper is technically sound and the data substantiates its claims and to provide perspective on the paper's significance to the field at large.
The editors also note that it's OK to say no if approached to be a peer reviewer. Being too busy to do a good job in the time given and having a conflict of interest are valid excuses to pass on reviewing, they say.