One of the greatest steps forward in science in the last several decades has been the widespread acceptance and use of the double-blind study. This has more to do with human nature than with the scientific study at hand, but science is done by people, so the statement holds. Direct experience of an event may not be as reliable as reasoning about evidence (scientific inference), even after the fact. This is a somewhat surprising notion, and one that has truly profound implications, not just for science, but for any field of testable knowledge.
A double-blind study is so called because neither the subjects nor the researchers administering the test are aware of the objective details of the test. For example, when testing a new drug, neither the patients nor the researchers recording the results know whether the patients are given the real drug being tested or a placebo. Such a study eliminates conscious and unconscious bias on both the parts of the subjects and researchers.
Another example is the analysis of the results of a particle physics experiment. Single-blind is sufficient in this case as the subjects (particles) are not capable of bias. One methodology is to subdivide the analysis into fractions the size of which are unknown to the analysts. They then would have no expectation of particular results for each fraction. Once the fractions have all been analyzed in this blind fashion, they can be assembled systematically and without bias into a complete (double-blind) study. There are several other methodologies used, each one carefully denying enough information to the analysts to ensure their biases are neutralized (knowledge of current theory, knowledge of apparatus used, knowledge of colleagues, etc).
This observer bias can have many causes, not the least of which is the pattern-seeking nature of the human mind. Our experience, expectations, and even desires can alter and augment our perceptions. In addition to observer bias, a well designed double-blind study can also minimize statistical illusions and false cause-effect conclusions.