Evaluation results and an autoalignment method for an optical head of a near-field recoding (NFR) system are presented. The focusing unit is an optical head of a NFR system and is composed of a solid immersion lens (SIL) and an objective lens (OL). Generally, the size of the focusing unit is smaller than that of the conventional optical recording head. Hence there are difficulties in assembling the small focusing unit precisely and a novel method for an effective assembly is required. We compose an evaluation system with an interferometer and evaluate some focusing unit samples aligned and assembled manually and present the obtained results. We also propose a conceptual method of autoalignment to assemble the focusing unit well by a pattern recognition using a neural network. Using the conventional optical tool, Code V, a tolerance analysis of the assembly error between the SIL and the objective lens and an interference pattern analysis for the assembly error are executed. Then, through an analysis of the simulation results, the autoalignment methodology using a neural network approach is proposed.