EA Backtesting and Curve Fitting Concern

I'm currently in the state of backtesting an EA and I'm using the optimize feature over three different variables. (Giving me a couple thousand combinations).

What I plan to do is run the optimization over 2003-2004 and then selecting those combinations I see working and running them again over each year 2005 and 2006. I'm looking for a somewhat consistant return over the 4 years, with acceptable drawdowns.

Is there a better way to do this? I considered having the optimizer run over the entire 4 year period, and then analyze those parameters which looked favorable to see if they gave the kind of returns and drawdowns I find acceptable for each year, however I wondered if I would be cureve fitting the data too much by doing that.

Any comments or suggestions?

I think it's important to have an 'out of sample' period to use after the EA has been optimized. I doubt that 2 years is required, but it's more a matter of preference.

I would optimize against 2 years worth of data leading up to August, then see how the EA performs after that based on the optimized settings. This is just my preference, I'm not a statistician.

Personally I only backtest the last 6 months, and then opimize for the last 6 months, and a final optimization geared towards the last month.

I dont believe in backtesting more than a years worth of data because in my humble opinion it would be too curve fitted.