Gerald M. Edelman, a 1972 Nobel laureate in Physiology/Medicine, in an interview with Istvan Hargittai:

Q: Have you tried to formulate a recipe for scientific discovery?
A: I'll tell you my theory of discovery by means of an anecdote. Beethoven's landlady says to him, "Beethoven, get out of my house. Your cat drinks my milk, you throw your laundry in the stairwell, and you pound on the piano all night, I can't sleep." And she laughs in response, "Ha-ha-ha-haaa" [the first strokes of the Fifth Symphony.] That's discovery, namely, contingency, accident, pattern, preconception, elaboration, and constantly playing back and forth against tradition of some kind; in Beethoven's case, it was Viennese classical music. In tracing a discovery there is an extraordinary complexity and diversity in the history, circumstance, cultural development, and technical skill. So to try to lay down any simple rule is not possible. (p 209).

It looks like there's no recipe for scientific discovery. On the other hand, if we start thinking about science as a system, in which discovery of a scientific discovery must be possible, we can take to the heart Edelman's words several pages below:

In terms of principle, selectional systems all have three constraints: (1) You must have a generator of diversity (G.O.D.). (2) You must have some mechanism for polling so that two domains which have no causal connection can sample each other with sufficient completeness. (3) You have to have some form of differentiated amplification, providing a selective advantage, for example, by producing enough antibodies of the right kind. (p 213).

What he is describing here is a system capable of solving detection problems, and then feeding the results of its work to a control mechanism that acts on the differences detected. Of course, in early stages of system development, diversity is not significant. But, as the system matures and diversity increases, the detection/control sub-system grows in importance because it provides for the greater efficiency of the top level system.
Success of Silicon Valley would be a great illustration to the interaction of the selectional principles. We have (1) a culture that encourage generation of diverse ideas; (2) various multi-stage talent detection mechanisms, e.g. startup funding and/or academic grant review processes; (3) a relatively large pool of VC money available for entrepreneurs.

References:

Candid Science: Conversations With Famous Biomedical Scientists
By Istvan Hargittai, Magdolna Hargittai
Published by Imperial College Press, 2006
ISBN 1860942881, 9781860942884
616 pages