multiply the std error given automatically by Prizm by the the result of the square root of 3 (or sample number) and you'll get the std deviation. It's going to be bigger, but it is the simplest way.

to make it simple SE= SD/ root of no of samples. So you can calculate back SD from this.

Thanks for your answers! But the problem is that I will have the IC50 value as concentration units and its standard deviation as log. How can I do to have standard deviation in concentration units?

Here the GraphPad FAQ explains how to have the SD not as log:

http://www.graphpad.com/support/faqid/1162/

Although if you read the Why? link, you'll see tat perhaps it's not a good idea.

You can find the answer here:

http://www.graphpad.com/support/faqid/1162/

Natalia: Did you find a way to do this...I am stuck in the same situation. The link above just talks about how to do this. Can we take anti-log of the standard error and then multiply by the square root of the sample size?

Do not use anti-log. As is described in the support information to Prism mentioned in the previous answer you should select an equation that contain EC50 rather than logEC50 and then "Prism will fit the EC50 or IC50 directly, with standard errors. "

Dose/concentration response curves on a log scale are log normally distributed, therefore transformation of the logIC50 values to molar will result in an asymmetrical distribution (see figure of randomly generated frequency distributions for log normally distributed IC50 values and those values converted to nM with a red line indicating the mean). This becomes a larger problem when you attempt to calculate the 95% confidence intervals from the transformed standard error for a molar IC50 as it can result in a lower limit that is negative, i.e. nonsensical. Use of log values or the negative log (pIC50) is much more appropriate and once you get accustomed to these, you will find that comparison of potencies is much easier and practical.