Using Laplace eigenvalue problem $$ -\Delta u_i = \lambda_i u_i \quad\mbox{in }\Omega, \qquad u_i = 0 \quad\mbox{on }\partial\Omega, $$ as a model problem, a generalization of error bounds from [1] to the case of tight clusters and multiple eigenvalues is presented. Individual eigenfunctions do not depend continuously on problem data, in general, and they are sensitive to small perturbation of the problem in the case of tight clusters and multiple eigenvalues. Therefore, we propose to estimate entire spaces of eigenfunctions corresponding to clusters and multiple eigenvalues. We derive a guaranteed and fully computable upper bound on a distance between the space of exact and the space of approximate eigenfunctions. The derived bound depends on the width of the cluster, the spectral gap between the last cluster and the following eigenvalues, and on the possible non-orthogonality of approximate spaces corresponding to the preceding clusters. The derived bound can be easily computed from two-sided bounds on eigenvalues and the approximate eigenfunctions. No flux reconstructions are needed. It is naturally evaluated recursively, starting from the lowest cluster. Numerical experiments compare several versions of the bound both in the energy and $L^2(\Omega)$-norms. Optimal rates of convergence are observed. More details are presented in [5]. An alternative approach [4] is a generalization of concepts from [2,3] to the case of clusters and multiple eigenvalues.