As you already know, you need to use a range of doses, preferably with replicates. But, you must observe a maximal or near maximal response. This may require more than one experiment as the dose range may need to be adjusted. Too many nearly maximal doses or no effect doses are not good. Perhaps a plot of dose vs response will produce what is called an 'S' shaped or nearly 'S' shaped curve when the doses are plotted on a logarithmic scale. Between the minimal and maximal doses the curve may have a region that appears to be nearly linear. The IC50 (ED50) is likely to be located an inflection point on the curve. Software, can be used for this type of analysis , check online. Some people have used EXCEL others have used R.

I agree with Mr Michael on this.