There is the ideal which explains why we need a pragmatic solution
For synoptic and climatological meteorology, the temperature required is a representative one of the 'free air' conditions over as wide an area surrounding the observing point as possible, with an internationally agreed height (for the thermometer bulbs, sensors etc.) of 1.25 m above local ground level. A fixed height must be specified, because vertical temperature gradients can be intense: for example on a clear, calm night or around the middle of the day with strong solar heating.
The best site for a screen, or thermometer shield for a land station is therefore over level ground, freely exposed to the sun and wind, but not sheltered by buildings, trees, bushes etc. The temperature sensor must be shielded from direct sunshine (hence a screen or shield) and precipitation (or a dry bulb becomes a wet bulb), and there must be a good circulation of air around the bulb/sensor head. If you have a garden, then the 1.25m above ground level can usually be met with ease. What it usually problematic is gaining sufficient clearance from adjacent buildings, trees etc.
The screen/shield should be positioned over grass (or less preferably, but still acceptable, loose soil), but not compacted soil, tarmac or concrete, as these media absorb and radiate solar energy strongly, and affect the readings quite significantly.
The bottom line is that the weather station should not be near a building - making it rather difficult to achieve if linked to a building management system
Pragmatically, there are three approaches of which one is reliable, one dreadful, and one relies on luck. First let's reject the "dreadful method".
Using your own local weather data
Just because sticking a weather temperature out of a window and calculating the degree-days using your Building Control System is easy AND VERY LOW COST - It does not make it cost effective.
In fact, though commonplace it is unbelievably stupid!!! - Why ? - When doing energy management we are seeking to validate decisions in a real context.If we look at the real context of "demand" as determined by a potentially broken system, we may see that software does "exactly what it should" given the inputs. We need to validate the inputs !
Using a free independent source of reliable weather data
This is the one that relies on luck. Maybe you live next to an air-field. Air fields and weather conditions have an obvious and important relationship.
Otherwise the typical distance between reliable weather stations is so great to make it meaningless for all but the broadest of analysis (eg. in the UK with a very much higher than average station density the typical spacing averages over 40km).
So if you live near one of the worlds Metar stations http://www.degreedays.net/ can help you (except if you believe that you building changes behaviour between day and night (many do)
Dr Russel Layberry provides some data (weekly resolution via teh Environmental Change Institute in Oxford
http://www.geog.ox.ac.uk/~rlayber/world_weekly_degree_day_data/
And in the UK data is available daily from 77 stations
http://www.eci.ox.ac.uk/research/energy/degreedays-weekly-daily.php
What we believe to be the right approach
If you look at a weather map, it has values for rain, pressure, wind and temperature for every point on the planet. These are calculated using super computers world-wide. Generally these provide the data against which local weather stations are vetted, and over time average zero statistical error or drift. Extracting data by Latitude, Longitude and ASl for a building in question is the approach we use. In general this is the best globally available reference data available. Our partner meteoblue AG of Basel Switzerland provide this to support our commercial client weather needs.
Update - Please note our free service based on public weather data