Not yet reviewed by Pith; the record is open.

We present 22\,901 OB spectra of 16\,032 stars identified from LAMOST DR5 dataset. A larger sample of OB candidates are firstly selected from the distributions in the spectral line indices space. Then all 22\,901 OB spectra are identified by manual inspection. Based on a sub-sample validation, we find that the completeness of the OB spectra reaches about $89\pm22$\% for the stars with spectral type earlier than B7, while around $57\pm16$\% B8--B9 stars are identified. The smaller completeness for late B stars is lead to the difficulty to discriminate them from A0--A1 type stars. The sub-classes of the OB samples are determined using the software package MKCLASS. With a careful validation using 646 sub-samples, we find that MKCLASS can give fairly reliable sub-types and luminosity class for most of the OB stars. The uncertainty of the spectral sub-type is around 1 sub-type and the uncertainty of the luminosity class is around 1 level. However, about 40\% of the OB stars are failed to be assigned to any class by MKCLASS and a few spectra are significantly misclassified by MKCLASS. This is likely because that the template spectra of MKCLASS are selected from nearby stars in the solar neighborhood, while the OB stars in this work are mostly located in the outer disk and may have lower metallicity. The rotation of the OB stars may also be responsible for the mis-classifications. Moreover, we find that the spectral and luminosity classes of the OB stars located in the Galactic latitude larger than 20$^\circ$ are substantially different with those located in latitude smaller than 20$^\circ$, which may either due to the observational selection effect or hint a different origin of the high Galactic latitude OB stars.