Dynamical systems are all around us, from various natural to industrial processes. Models are developed for these processes, which are then used to simulate the behavior of processes and/or to control over them. In order to increase the precision, there are more complicated models which can’t be effectively used. It is therefore necessary to look for reduced models, models that are simpler, but close enough to complicated models. Linear, time-invariant, continuous-time, finite dimensional, real, stable systems of the form \(Ex'(t) = Ax(t) + Bu(t), y(t) = Cx(t)\), are observed, where \(E, A \in \mathbb{R}^{n \times n}, B \in \mathbb{R}^{n \times m}\) and \(C \in \mathbb{R}^{p \times n}\) are constant matrices, and \(x(t) \in \mathbb{R}^n, u(t) \in \mathbb{R}^m\) and \(y(t) \in \mathbb{R}^p\) are the state, input and output of the system. This system is said to be \(n\)-dimensional. The goal of model reduction is, for given \(n\)-dimensional system, to find an \(r\)-dimensional system (where \(r \ll n\) is also given), such that the output of the reduced model is as close as possible to the output of the full model. The meaning of closeness which is used in this work is that which is given by the norm in the Hardy’s space \(\mathcal{H}_2^{p \times m}(\mathbb{C}_+)\). This work presents a method that finds a reduced model satisfying some necessary conditions of optimality in \(\mathcal{H}_2\) norm, given in the form of tangential Hermite interpolation. It also presents a method based on Loewner’s matrices which does not depend on any realization (\(E, A, B, C\)) and is applicable to infinite dimensional systems. All methods have been tested on examples from Oberwolfach and NICONET databases and the results are compared with the results obtained using balanced truncation.