LiDAR point information is usually available as a set of ASCII or LAS data files:

ArcGIS only supports LAS data files; to use ASCII LiDAR data with Arc, you’ll need to use an external tool to convert to LAS.

LAS files cannot be added directly; they must be combined into an LAS dataset that sets a consistent symbology and spatial reference for the entire collection.  To create a LAS dataset, go to ArcCatalog and right-click the folder you want to store the dataset in, and select new>LAS Dataset:

Note that you will need 3D Analyst or Spatial Analyst activated to do this.  I recommend checking all the extensions to be sure your tools run the first time.

Right-click the dataset in ArcCatalog and choose Properties.

Ensure your dataset has an appropriate coordinate system in the XY Coordinate System tab, which you’ll need to get from the metadata for the LiDAR.  Next, activate the LAS Files tab and click Add Files or Add Folders.   Once you are done adding files, activate the Statistics tab and press the Calculate button.

At this point, you can import your data into either ArcMap or ArcScene.  There are pros and cons to both.  As far as I’ve been able to determine, it is impossible to plot 3D point clouds in ArcMap with a DEM or map base.  This is possible in ArcScene, and it is also possible to color points according to intensity in 3D view, but unlike in ArcMap there is no ability to adjust point size, and very limited ability to adjust colors of points, at least as far as I’ve been able to determine over the last few days.

Some LAS datasets will include RGB information in addition to intensity, which allows 3D true-color visualizations.

This image shows the Middle Lookout Creek site in the HJ Andrews Experimental Forest as a point cloud colored by intensity in ArcScene.  The creek and some log jams are visible on the right, and a road is visible on the left.

To convert LiDAR points to a DEM, you’ll need to convert the dataset to a multipoint feature class first.  Open ArcToolbox and go to 3D Analyst Tools> Conversion>From File>LAS to Multipoint.  Select the specific LAS files you want.  It’s likely that you’ll want to use the ground points only.  The usual class code for ground points is 2, but you’ll want to check this by coloring the points by class.

Once you’ve created the multipoint feature, you need to interpolate the values between the points.  There are several ways to do this, with advantages and disadvantages.  Popular methods are Spline and Kriging.  Spline generates a smooth surface but can create nonexistent features, and doesn’t handle large variations over shorter than average distances very well.  Kriging is hard to get right without experience in what the parameters do, and can take a long time to achieve the best results, but attempts to take spatial autocorrelation into account.  In general, Kriging is better for relatively flat areas, and Spline is better for sloped areas.  Inverse Distance Weighting is popular, but produces results similar to a very poorly configured (for terrain anyway) Kriging interpolation.  I find that a safe but time-consuming bet is Emprical Bayesian Kriging, which can be found in the toolbox under Geostatistical Analyst Tools> Interpolation, along with a few other advanced interpolation methods that I am not as experienced with.  If anyone else is familiar with these, I’d welcome a post explaining how best to use them.