Understory trees often emerging beneath dominant tree layer in even-aged stands have significance for timber harvesting operations, forest regeneration, landscape and visibility analysis, biodiversity and carbon balance. Airborne laser scanning (ALS) has proven to be an efficient remote sensing method in inventory of mature forest stands. Recent introduction of ALS to operational forest inventory systems could potentially enable cost-efficient acquisition of information on understory tree layer. In this study, accurate field reference and discrete return LiDAR data (1–2 km flying altitude, 0.9–9.7 pulses m-2) were used. The LiDAR data were obtained with Optech ALTM3100 and Leica ALS50-II sensors. The field reference plots represented typical commercially managed, even-aged pine stands in different developmental stages. Aims of the study were 1) to study the LiDAR signal from understory trees at pulse level and the factors affecting the signal, and 2) to explore what is the explanatory power of area-based LiDAR features in predicting the properties of understory tree layer. Special attention was paid in studying the effect of transmission losses to upper canopy layers on the obtained signal and possibilities to make compensations for transmission losses to the LiDAR return intensity. Differences in intensity between understory tree species were small and varied between data sets. Thus, intensity is of little use in tree species classification. Transmission losses increased noise in intensity observations from understory tree layer. Compensations for transmission losses were made to the 2nd and 3rd return data. The compensations decreased intensity variation within targets and improved classification accuracy between targets. In classification between ground and most abundant understory tree species using 2nd return data, overall classification accuracies were 49.2–54.9 % and 57.3–62.0 %, and kappa values 0.03–0.13 and 0.10–0.22, before and after compensations, respectively. The classification accuracy improved also in 3rd return data. The most important variable explaining the transmission losses was the intensity from previous echoes and pulse intersection geometry with upper canopy layer had a minor effect. The probability of getting an echo from an understory tree was studied, and differences between tree species were observed. Spruce produced an echo with a greater probability than broadleaved trees. If the pulse was subject to transmission losses, the differences were increased. The results imply that area-based LiDAR height distribution metrics could depend on tree species. There were differences in intensity data between sensors, which are a problem if multiple LiDAR data sets are used in inventory systems. Also the echo probabilities differed between sensors, which caused minor changes in LiDAR height distribution metrics. Area-based predictors for stem number and mean height of understory trees were detected if trees with height < 1 m were not included. In general, predictions for stem number were more accurate than for mean height. Explanatory power of the studied features did not markedly decrease with decreasing pulse density, which is important for practical applications. Proportion of broadleaved trees could not be predicted. As a conclusion, discrete return LiDAR data could be utilized e.g. in detecting the need for initial clearings before harvesting operations. However, accurate characterization of understory trees (e.g. detection of tree species) or detection of the smallest seedlings seems to be out of reach. Additional research is needed to generalize the results to different forests.