Explanation of: "You will be fitting to the actual data. Even though the graph has a log axis, you will not be fitting to log values."

Dose response curves are commonly shown with the concentration or dose shown on a log scale. Shown this way, the sigmoidal curve is symmetrical. It also makes sense because the doses or concentrations in these kinds of experiments are usually equally spaced on a logarithmic scale. 

There are two ways to handle this situation. 

• Enter the concentrations as logarithms, or use Prism's Transform analysis to transform X values entered as doses or concentrations to their logarithms. Then fit models (equations) written to expect X to be log(concentration) or log(dose). 
• Enter the concentrations or doses directly, then stretch the X axis to a log scale. Prism can do this with a control at the upper right of the Format Axis dialog. Then fit the curve. 

All the dose-response curves built in to Prism assume that X is log(dose) or log(concentration). If you enter concenrations or doses (and don't transform) then any fit of these models will produce unhelpful results. Even if you stretch the axis to a log scale. Prism sees the actual numbers in the table you chose as input to the analysis. Prism's nonlinear (or linear) regression does not take into account the fact that you have stretched the axis to a log scale. Nonlinear (and linear) regression only see the data (or transform results) table, and don't "see" the graph at all. 

It is possible (even easy) to write the dose-response models to expect X to be concentration, rather than log(concentration), but these are not built-in to Prism. 

Even if you enter data as logarithms, or transform to logarithms and plot those, Prism can make the graph attractive. Instead of labeling the axis with logarithms (-3), Prism can label with antilogarithms (0.001) or powers of ten (10^-3). Choose this in the Format Axes dialog.