...and if you're using data from a noisy source (i.e. experimental tomography data) there are *much* better ways to generate a surface than thresholding by intensity. (If you only threshold, then generally your interpolating functions need to be dramatically underfitted to avoid surface artifacts.)
This process is known as 'segmentation' and there's a lot published about it. An old reference is appended to get you started.
Physica A: Statistical Mechanics and its Applications
Volume 339, Issues 1