Dendrobium is a commonly used Chinese herbal medicine, often using fresh or dry stems into the medicine, beneficial to the stomach, nourishing yin and clearing heat. In recent years, pharmacological studies have found that Dendrobium has the functions of anti-cataract, anti-oxidation, anti-tumor and improving immunity. It has remarkable effects in many cases, which has attracted the attention of scholars at domestic and abroad. However, the contents of amino acids and trace elements in Dendrobium collected at different times are different, and their medicinal value and price are different. So the study of price grade discrimination of Dendrobium is of great significance. In order to quickly identify Dendrobium with different price and efficacy, the random forest classification modela combined with laser induced breakdown spectroscopy (Laser-induced Breakdown Spectroscopy, LIBS) was developed to model the price grade of Dendrobium. In this paper, five samples of Dendrobium were selected for modeling. In order to analyze the samples accurately and stably, all Dendrobium samples were pressed to reduce the experimental error. The Nd∶YAG pulse laser with 1 064 nm wavelength was used as the excitation light source, the detection delay of 50 mJ, laser pulse energy was set to 1 μs, the spectral data of five grades of Dendrobium were collected, 40 sets of spectra were collected from each grade of samples, and a total of 200 sets of data were collected. Normalized processing was used to convert all spectral data from -1 to 1. The principal component analysis (PCA) was used to analyze the normalized spectral data. The score matrix of the first seven principal components was obtained by principal component analysis, and the cumulative interpretation of the total spectral information was 95.24%. So seven principal components were selected as input, and a random forest identification model with 220～880 nm was established. The number of Dendrobium samples was disrupted, and 50% spectral data were randomly selected as training sets, and 50% spectral data were left as test sets. The default number of decision trees (ntree) was 500, and the number of attributes in the split attribute set (mtry) was 5. The model was established to classify Dendrobium in different grades. And the recognition rates of grades one, two, three, four and five were 95.45%, 100%, 78.26%, respectively. 94.12%, 85%, with an average recognition rate of 90.57%. In order to improve the recognition rate, the influence of different ntree and mtry on the classification model was studied, and the two parameters of the random forest were optimized by using the out-of-bag data error rate estimation. The ntree was 300, the mtry was 1, the recognition rates of grade one, two, three, four and five were 100%, 100%, 92.31%, 100%, 90%, the average recognition rate was 96.46%, and the recognition rate was increased by 5.89%. In conclusion, it is feasible to identify the Dendrobium classification by LIBS technology combined with the optimized random forest model, which provides a feasible discrimination system for the rapid identification of Dendrobium classification with different prices in the future.