Science is in the method. Observation->Hypothesis->Experiment, at a high level. The key bits are really Hypothesis and Experiment.
A hypothesis must be:
1. Testable. There must be a way to design an experiment that can falsify the hypothesis. This is often the argument brought up against string theories. But this testing doesn't have to be practical to make it scientific, just possible. The work being done on these is to find practical options to test. And contrary to some statements, there's not that much money spent on these. A lot of money is spent on basic research, but that research is MUCH broader that just string theories.
2. Explanatory. The hypothesis should explain the observed facts. This includes both new and old facts...the hypothesis shouldn't just explain whatever new phenomena it's looking at, but also be consistent with past experiments and data. Note that I did not say past theories...it may overturn a current theory, but it must still adequately explain the data that led to that earlier theory. For example, Einstein didn't prove Newton wrong, so much as show that Newton's laws were an approximation of General Relativity that applied at the "usual" scales of force and speed. GR is perfectly consistent with Newton, for most day-to-day purposes.
Experiment is the other big one. Again, to be scientific, an experiment must be:
1. Specific. The variables involved need to be isolated, so that results can be correctly interpreted. This delves into large areas of logic, statistics, and so on, so there is no easy, simple answer. But this is what leads to things like placebo-controlled double-blind trials for medical interventions. On a basic level, everyone should be doing this in high-school science. When you're testing acceleration of gravity in high school physics, for example, they often have you do something like running a weighted sled down a ramp. Parts of the procedure are obviously there to remove extra variables: using calibrated weights, verifying the ramp angle is identical between runs, using the same sled each time, starting the ramp from the same location, and so forth. All of this is to try and isolate just the variables you're interested in. If you randomly set the ramp, guestimate weights, and so forth, there's no way to tell what variable may have caused the differences your seeing. This is where Sagan's Dragon in the Garage comes in to play with pseudo science, and one of the big indicators. Well-designed, failed tests, instead of being accepted, will be explained away with new variables that were not considered before the experiment. That amounts to a new hypothesis, one which that experiment cannot support. A failed homeopathy experiment, for example, can't then be used as proof of energy vibrations from fluorescent lights, say, if that wasn't a variable initially controlled in the experiment. At best, that failed experiment becomes the observation for the development of a new hypothesis, which then must be tested again. The best you can get is the current experimental results tossed out.
2. Rigorous. The results must be calculated and evaluated honestly. This ties into the last bit of the above, but is worth a mention separately too.
3. Honest. In this, I meant the experiment must be designed so that falsification of the hypothesis is possible. Ties into 1 and 2 as well, but wanted to mention it here. To borrow from an analogy I used to use:
Science is not like a job interview, or a persuasive speech. The goal isn't (or shouldn't be) to succeed. The goal is to find out what works and what doesn't. I like to compare it to vehicular safety testing. When you're testing a vehicle for crash safety, the goal isn't to declare the vehicle as safe...it's to find out if it's safe, and how safe it is. The same thing holds true for a scientific hypothesis or theory. When testing front-end collision safety, you don't put 4 feet of foam rubber in front of the brick wall the car is going to hit, so as not to hurt it....you speed the car up to 60mph and slam it into brick head-first. The way the car (the hypothesis) fails tells you more about safety, about what works and what doesn't. In some cases, General Relativity being a case in point, the theory is so well tested that we're more often seeing the wall fail then the car.
But this highlights another place where real science separates from pseudoscience. If you actually study it, pretty much every expert will tell you that GR is NOT sacrosanct. We know, for a fact, that it is at best incomplete. I hesitate to use the word "wrong", because it's more like the case of GR replacing Newton. It wasn't that Newton was wrong, just that it was an approximation within certain limits. The same is true of General Relativity, and science readily admits that.
— that I find interesting but I haven't got the time to reply to all comments, I beg you all excuse me for this.