Biological systems such as axonal growth cones perform chemotaxis at micrometre-level length scales, where chemotactic molecules are sparse. Such systems lie outside the range of validity of existing models, which assume smoothly varying chemical gradients. We investigate the effect of introducing discrete chemoattractant molecules by constructing a minimal dynamical model consisting of a chemotactic cell without internal memory. Significant differences are found in the behaviour of the cell as the chemical gradient is changed from smoothly varying to discrete, including the emergence of a homing radius beyond which chemotaxis is not reliably performed.