Pictograms are basically designed as simplified images of human behavior or activities. It is a global and intuitive communication tool that does not require language or background knowledge. On the other hand, recent years have seen the pictograms’ creation and installation that give priority to uniqueness of design or affinity for the atmosphere of the place. In these cases, the pictograms are difficult to convey information obviously and intuitively, and do not function as ‘pictograms’ in the first place. This has resulted in a situation where, for example, additional text is added to the pictogram in the production process, or information is added after installation by stickers or additional. This is a “Design Fails” case, In this study, we design an Image Recognition application that extract pictograms that immediately and intuitively provide information and it learns the data. Based on the training data, the system judges whether the target pictogram functions as ‘pictograms,’ quantify its relative appropriateness, and also learns the results. It is expected that this application can last-check whether pictograms designed or installed are informative enough for everyone, to be clear at a glance.