In the present work, the combination of chemical immobilization with electron beam lithography enables the production of sensitive and reproducible SERS-active areas composed of stochastic arrangements of gold nanoparticles. The number of nanoparticles was varied from 2 to 500. Thereby a systematic analysis of these SERS-active areas allows us to study SERS efficiency as a function of the number of nanoparticles. We found that the experimental parameters are critical, in particular the size of the SERS-active area must be comparable to the effective area of excitation to obtained reproducible SERS measurements. The sensitivity has also been studied by deducing the number of NPs that generate the enhancement. With this approach we demonstrates that the maximum enhancement, the best sensitivity, is obtained with the smallest number of nanoparticles that is resonant at a given excitation wavelength.