In a recent post I expounded my views on the need of eliminating, or greatly reducing the amount of required course work for science graduate students. Fundamentally, the idea is that students are going to learn mostly by doing research, and that course work largely gets in the way.
That is not to say that there is no room for courses in graduate school, though. However, the style and format of courses that I support offering at the graduate level should be markedly different than that of regular undergraduate courses. I like to refer to them as short courses.
A short course should be first and foremost, well, short. The duration of a typical one-semester course  at an American university is between 13 and 16 weeks, with 150 minutes worth of lectures every week. A short course should last between four and six weeks, with class meeting on average at the same frequency as for a regular course . The idea is that students ought not be required to commit to attend lectures through a period of time (an entire semester) whose length may render it impractical. This is especially important for advanced graduate students, who will spend the greatest part of their time focusing on their dissertation projects, traveling to conferences or job interviews, and writing their thesis .
More generally, a short course should really aim at being as little disruptive as possible to graduate student research schedule. Instructor should keep this in mind before assigning homework, requiring students to write essays or deliver in-class presentations, or otherwise setting unrealistic goals that would require students to set aside a significant amount of time. This would defeat the purpose of the whole exercise.
Secondly, it should not be designed as a foundation course, including an overview of a (possibly broad) field of research, going through basic derivations, reviewing the literature to a significant extent. Rather, it should be very narrowly focused, with the goal of providing students with practically utilizable knowledge, ideally profitably applicable to their own research. The first example that comes to mind is a course on data analysis and error estimation, clearly of interest for students engaged in experimental research, or in theoretical work making use of numerical simulations (which, as we all know, is not really theoretical work) .
In other cases, course audience might also include students whose area of research is not the one covered by the course, but who may nonetheless benefit from knowing its basic ideas. For example, students may be interested in understanding how a specific experimental or theoretical technique works, not so much because they themselves may be planning to make use of it, but because of its popularity among scientists working in their area of research. While they may not seek to becoming experts, a working knowledge of that tool, of its advantages and limitations, is often quite valuable .
Offering short courses can be a good ideas from a very practical standpoint as well. While it may be difficult to convince a resident faculty to teach one, given the significant amount of preparation work required and the presumably limited teaching credit that university administrators are going to be willing to grant, a visiting professor may jump at the opportunity, as a short course is often an excellent recruiting tool.
In practical terms, this will allow a department or a college to take advantage of the presence on campus of a scientist whose expertise in a specific field may be “unique”, not matched by any member of the faculty at that institution; moreover, the fact that this person will be engaged in some form of teaching, albeit in the form of a short course, may make it possible to use monies allocated for teaching to fund his/her visit.
What are then the problems ? Mostly the opposition of faculty who support instead a more “traditional” approach to graduate courses (tying up students for thirteen weeks, assigning homework problems that will keep them up at night, final exams… all that good stuff).
 Some campuses work on a quarter system. The duration of a course in that case is typically around ten weeks.
 Obviously, the instructor of a short course should be given the flexibility of organizing the course in a way that best suits the need of the students. For example, as graduate students travel frequently, it may be occasionally opportune to spread the course over a period of time longer than six weeks, or to concentrate some of the lecturing during a shorter (e.g., two weeks) time frame. As attendance of short courses is typically small, instructor will usually not find it too difficult to select lecturing time to accommodate all the student requests.
 Short courses can clearly benefit postdocs too, for whom the same considerations apply.
 Although science graduate students should be familiar with a subject of that type, having learned it in undergraduate physics and statistics courses, in practice they often need a refresher. Moreover, undergraduate foundation courses seldom give students practical recipes, which are really what graduate students need, for the most part.
 Case in point: this term I am teaching a course on Monte Carlo methods applied to problems of Statistical Physics problems. The course is attended mostly by students and postdocs who are not practitioners of Monte Carlo, not are they planning to become any time soon, but are curious to know what it is all about, in order to be able to assess critically by themselves calculations done by others.