We suggest some new methods of sparse recovery in deterministic and statistical models. Examples include problems with missing data, con- strained minimization and other. We prove sparsity oracle inequalities both for the case of exact sparse model and for approximately sparse solutions.