Context. The EROS-2 survey has produced a database of millions of time series from stars monitored for more than six years, allowing us to classify some of their sources into different variable star types. Among the so-called, long period variables (LPVs), which are known to follow particular sequences in the period-luminosity diagram, we find long secondary period variables whose variability origin remains a matter of debate. Aims. We analyze data for the 856 864 variable stars that have been discovered in the Large Magellanic Cloud (LMC) that are present in the EROS-2 database, to detect, classify, and characterize LPVs. Methods. Our method for identifying LPVs is based on the statistical Abbe test. It investigates the regularity of the light curve with respect to the survey duration in order to extract candidates with long-term variability. The period search is performed using Deeming, Lomb-Scargle, and generalized Lomb-Scargle methods, combined with a Fourier series fit. Color−magnitude, period-magnitude, and period-amplitude diagrams are used to characterize our candidates.Results. We present a catalog of 43 551 LPV candidates for the Large Magellanic Cloud. For each of them, we provide up to five periods, a mean magnitude in EROS-2, 2MASS, and Spitzer bands, B E −R E color, R E amplitude, and spectral type. We use infrared data to distinguish between RGB, O-rich, C-rich, and extreme AGB stars. Properties of our LPV candidates are investigated by analyzing period-luminosity and period-amplitude diagrams.