The recognition rate of holographic neural synapses, performing a pattern recognition task, is significantly higher when applied to natural, rather than artificial, images. This shortcoming of artificial images can be largely compensated for, if noise is added to the input pattern. The effect is the result of a trade-off between optimal representation of the stimulus (for which noise is favorable) and keeping as much as possible of the stimulus-specific information (for which noise is detrimental). The observed mechanism may play a prominent role for simple biological sensors.