Accurate detection and classification of primary cilia in microscopy images is an essential and fundamental task for many biological studies including diagnosis of primary ciliary dyskinesia. Manual detection and classification of individual primary cilia by visual inspection is time consuming, and prone to induce subjective bias. However, automation of this process is challenging as well, due to clutter, bleed-through, imaging noise, and the similar characteristics of the non-cilia candidates present within the image. We propose a convolutional neural random forest classifier that combines a convolutional neural network with random decision forests to classify the primary cilia in fluorescence microscopy images. We compare the performance of the proposed classifier with that of an unsupervised k-means classifier and a supervised multi-layer perceptron classifier on real data consisting of 8 representative cilia images, containing more than 2300 primary cilia using precision/recall rates, ROC curves, AUC, and Fβ-score for classification accuracy. Results show that our proposed classifier achieves better classification accuracy.