Abstract. We present a new approach to estimate local Lyapunov vectors. The so called self-breeding method is based on the breeding of growing modes technique from medium range weather forecasting and consists of a continuous forecasting and rescaling cycle. Using the Lorenz96 model we test and characterize the behavior of the algorithm regarding error growth, spatial perturbation structure estimates and orthogonalization. The results indicate that the method can be used to generate error growing modes optimized for a certain rescaling interval, thus enabling the user to target specific scales of error growth. When an additional orthogonalization procedure is applied, the method is able to produce structures representing the error growth subspace spanned by the largest Lyapunov vectors.