I have no experience with MS programs. The following is for American doctoral programs.

The applicant should have a decent command of spoken English. Communication is important, in class, in research discussions, and in the likely TA duty. 

The applicant needs to have a solid undergraduate background, preferably with a few graduate courses thrown in. Some graduate programs are more forgiving of inadequate backgrounds than others, but the fact is that the requirements for graduating with a mathematics major in many US schools are far below what is needed to begin graduate study. If a candidate has not taken some graduate classes in mathematics, there should be some good explanation for this, e.g, that the student changed from physics to mathematics. 

There should be some indication that the applicant knows what research is, and that the point of graduate study is research. It is to be expected that most attempts at undergraduate research in mathematics do not produce meaningful results. 

The applicant should have an area of interest which coincides with that of a professor in the department who is taking more students. This is a more important issue in smaller departments. 

High GRE scores are expected. Their absence is an indicator that something is wrong. 

It's a plus if the applicant has shown some level of dedication to some area which might not be related to mathematics, but shows discipline and perseverance. Programming skills are a plus. 

Outside funding is rare, but a plus.

I'm skeptical of the use of Putnam scores. I say that as someone who had decent Putnam scores and who coached a top 5 team. Putnam performance is correlated with ability to solve puzzles rapidly and practice. It's not as well correlated with research ability.