Depends on what you're backtesting and how you're backtesting it.
Usually, when people refer to "backtesting", they mean plugging various indicators or whatever patterns can be coded into a computer program and being presented with seemingly important results. Many people consider this exercise to be meaningless and pointless.
However, if you're backtesting in order to define a setup based on S/R or to test the viability of an S/R setup you've already defined, you're going to have to actually look at the charts, by hand, personally. Sometimes you have to look at only a dozen to determine that what you thought was a great setup is really no better than a coin flip. Or you might be encouraged and go on to do a dozen more. Depends.
But if you're backtesting something specific, such as how to trade around Fed announcements, then you select those days on which Fed announcements occur, along with perhaps the two or three days preceding those announcements, and do your backtest using only those charts, going back for as many of these "sets" as you like. Similarly, if you're currently in a trading range, even a broad trading range, then you want to conduct your backtest using those periods that were also in a trading range (backtesting something using charts from a trending market isn't going to tell you much, except perhaps the wrong thing). For example if you were backtesting something in the NQ for current use, just about anywhere in the period from late November '05 to early May '06 would be useful. And that's a lot of charts.
I suppose one could have only one strategy for any market, but why, especially if one can have one primary strategy for, say, non-trending markets, with what might be only a relatively minor modification to alter it to a strategy for trending markets, which might entail nothing more dramatic than changing the stops one uses and where he places them.
