Traditional deep neural network classifier receives input data and passes through hidden layers to output predicted labels. In this paper, I propose an Inverted Conditional Generator Classifier that uses conditional generators to find a pair of condition vector and latent vector that can generate the data closest to the input data, and predict the label of the input data. The conditional generator is a generative model that receives latent vector and condition vector, and generates data with desired conditions. A decoder of conditional VAE [1] or a generator of conditional GAN [2] can be a conditional generator. The inverted Conditional Generator Classifier uses a trained conditional generator as it is. The inverted conditional generator classifier repeatedly performs gradient descent by taking the latent vector for each condition as a variable and the model parameter as a constant to find the data closest to the input data. Then, among the data generated for each condition, the condition vector of the data closest to the input data becomes the predicted label. Inverted Conditional Generator Classifier is slow to predict because prediction is based on gradient descent, but has high accuracy and is very robust against adversarial attacks [3] such as noise. In addition, the Inverted Conditional Generator Classifier can measure the degree of out-of-class through the difference between the generated nearest data and input data. A high degree of out-of-class means that the input data is separate from the cluster of each class, or Inverted Conditional Generator Classifier has little confidence in prediction. Through this, Inverted Conditional Generator Classifier can classify the input data as out-of-class or defer classification due to the lack of confidence in prediction.