I usually try to fit the formula with unknown parameters on the experimental points by using lsqnonlin fonction in MATLAB (genetic algorithm).

As a optimisation parameter, you could have axis offset, scale and exponent for exemple.

If a sigmoid function has the shape y = a + b/[ 1 + exp (-c(x-x0)) ], then the inverse function is simply x = x0 + (1/c)*log [(y-a)/(y-b-a)].  Fitted parameters are x0, a, b and c.  Start with a=35, b=15.

Hope it helps

Finney's Probit analysis is the proper equation:

http://userwww.sfsu.edu/efc/classes/biol710/probit/ProbitAnalysis.pdf

@ Marek Wojciech Gutowski, Previously I tried x=x0-(1/c)*log(b/(y-a)-1). I will try the equation you suggested.

Thank you.

@ Luisian Cundin I will read the analysis and try to understand it.

Thank you.