This answer may be a little more than you wanted!

The key and essential parts of any scientific experiment are reasonably simple to state, but often difficult to carry out.

The first, and hardest part is to frame a question that is both unambiguously answerable and significant. This is generally not done 'out of the blue' but requires thoughtful consideration and knowledge of the system under study.

Research is necessary to discover what is already known about the system and what remains to be known. Research is done primarily by reading the published work of others and thinking about their observations.

Once you are familiar with a system, you may be able to form a working hypothesis. This is an assumption of how the system works. It is not really important whether it is right or wrong, but whether it is logical and can be tested.

Experiments generally take one of two forms: observational or manipulative.

In an observational experiment, such as is common in astronomy, your hypothesis enables you to make a prediction. Look at this region of the sky with this type of instrument and you will make the following measurement.

The more accurate your predication, the more reason you will have to believe that your original hypothesis was correct. The more unexpected your prediction, the more impact it will have.

If your observations do not agree with your predictions, then either there is something wrong with your hypothesis or with your measurements. You perform controls (see below) to determine which is the case.

In a manipulative experiment, you attempt to perturb the system is a specific manner. Based on your hypothesis, you predict that such a perturbation will produce a specific response in the system. You use instruments to determine whether the system shows the expected behavior.

Again controls are used determine whether you actually perturbed the system in the manner you thought you did and whether the response of the system was specific to the perturbation you introduced.

The more accurately you predict the response of the system, and the more accurately your measurements conform that the prediction, the more support your hypothesis receives.

If the system does not respond as your hypothesis predicts, either you did not actually do what you thought you did, you did not accurately measure the system's response, or your hypothesis does not accurately describe the system.

Controls come in two types. Positive control experiments determine whether your manipulation of the system produced the effect you expect. For example, assume you want to test whether an specific antibiotic kills a newly identified type of organism. An important positive control would be confirm that your sample of the antibiotic is active, that it can kill organisms you already know are sensitive to it. In an observational experiment, a positive control would determine that the instrument you are using produces accurate and reproducible measurements. This is typically done using "standards", samples whose level of the quality to be measured is known.

Negative control experiments try to isolate the change in the system to a single variable. Suppose you are using an chemical in your experiments. The chemical must be dissolved in ethanol and then applied to the skin. A negative control experiment, would test whether applying ethanol alone to the skin leads to the same effect as ethanol plus drug.

In studies of drug effects on humans one also has to take into account the often subconscious perceptions of both patients and researchers. If, for example, a person is feeling depressed you only know whether a drug that you think will acts as an anti-depressant (your hypothesis) actually works by asking or observing the patient. If the patient thinks they are getting an effective drug, they may feel better whether the drug is effective or not - this is known as the placebo effect. If the researcher thinks that the drug should make the patient feel better, that preconception may bias their evaluation of patient.

To control for these effects, the most accurate drug experiments are performed "double-blind". Neither the researcher nor the patient receiving the drug know whether they are actually getting the drug to be tested or a inert substitute (a placebo). To work, the placebo needs to be made to look exactly like the drug -- if the drug is a blue and triangular pill, so is the placebo. Only after the study is over is the data decoded to determine which patients received the drug and which the placebo.

In medicine controls are critical and sometimes problematic. Imagine a surgical procedure that has been proposed to cure a particular disease (the hypothesis). To determine whether it actually works, a neurosurgeon must opens the skull and perform the procedure. But what if just opening the skull produces the same effect? To test this, a group of patient's should be subjected to a 'mock' version of the operation - their skull will be opened, left open for the time it take to perform the procedure, but no procedure will actually be performed.

In the end, the validity of a particular experiment depends upon the controls performed, the accuracy of the hypothesis' predications and the accuracy of the measurements made. If everythingf is done well, the hypothesis will either be confirmed, revised or rejected and new hypothesis and predictions made and tested.