diff --git a/README.md b/README.md
index 02213bc..5b97563 100644
--- a/README.md
+++ b/README.md
@@ -2,6 +2,10 @@
 
 ## Introduction
 
+Bicycle Network Analysis (BNA) is a way to measure how effectively people
+can connect to destinations via bike on comfortable, "low-stress" routes.
+More info on the methodology at the [What is the BNA?](what_is_bna.md) page.
+
 Python module to implement BNA logic on a PostGIS database. Uses a configuration
 file for customizability. The current implementation requires a user to run the
 old-style BNA on the command line to import datasets. This library can build the
diff --git a/what_is_bna.md b/what_is_bna.md
new file mode 100644
index 0000000..930e8a9
--- /dev/null
+++ b/what_is_bna.md
@@ -0,0 +1,140 @@
+# What is the BNA?
+
+Network connectivity can be a difficult concept to describe, understand, and --
+crucially -- to measure. Traditionally, cities attempting to quantify the
+usefulness of their bike networks have fallen back on easily-measured attributes
+like the mileage of bike lanes, or an as-the-crow flies distance to the nearest
+bike facility.
+
+Though there may be some correlation between "bike friendliness" and these crude
+measures, they fail to capture the importance of having an _interconnected_
+network of comfortable bike routes, in addition to the role that _destinations_
+play in connecting people to places. The BNA aims to capture the importance of
+the interconnectedness of bicycle routes by measuring access to destinations.
+
+There are three steps to the BNA (not counting the process of assembling the
+necessary input data as described in the [import instructions](import.md)).
+
+1. Level of Traffic Stress
+2. Block-to-block connectivity analysis
+3. Aggregation of destinations
+
+## Level of Traffic Stress
+
+Level of Traffic Stress (LTS) is a concept first described by Peter Furth, Maaza
+Mekuria, and Hilary Nixon in a paper published by the Mineta Transportation
+Institute at San Jose State University. In the paper, they set forth roadway
+conditions that contribute to cyclist stress. These are things you'd probably
+expect: high traffic speeds are uncomfortable for bicycling, quiet residential
+streets are less stressful than busy arterial roadways,etc. The original paper
+is published
+[here](https://transweb.sjsu.edu/research/low-stress-bicycling-and-network-connectivity).
+
+One important aspect discussed in the paper is the critical role intersection
+crossings play in route continuity. It's easy to forget the importance of
+intersections when looking at a map of bicycle routes because the intersections
+themselves take virtually no space. Contrary to the visuals, intersections
+actually have an outsize importance in terms of routes a typical cyclist might
+feel comfortable following. If you've ever biked down a bucolic residential
+street and found you had to turn back at the virtual wall of high-speed traffic
+zooming down the arterial at the end of the block, you understand this concept
+well.
+
+Another significant topic discussed in the paper is the principle of the
+"weakest link". This is an assumption that, in the minds of most cyclists, the
+highest stress level encountered on a route defines the stress level of the
+entire route. In the words of the authors, _"If people will not use links whose
+stress exceeds their tolerance, **several low-stress links cannot compensate for
+one high-stress link**."_ The implication of this is significant for network
+planning as the prevailing attitude has been to provide comfortable routes where
+feasible but accept some compromises in constrained locations (such as busy
+intersections). This is usually done in the hope that users will stomach a
+short, uncomfortable segment as long as the rest of the route is pleasant.
+
+The BNA adopts the general concepts of LTS for rating the comfort of possible
+route options. The lookup tables used as a basis for the BNA are published on
+Peter Furth's academic website
+[here](http://www.northeastern.edu/peter.furth/criteria-for-level-of-traffic-stress/).
+There are some deviations from Dr Furth's tables, including for features not
+covered in his work, such as HAWK signals. We've also made some adjustments at
+the margins to better fit conditions as we've observed them doing bike network
+planning around North America.
+
+The main magic of using LTS in the BNA is that it elevates the importance of
+intersections and insists on a continuous, comfortable route. Fortunately, the
+BNA is flexible enough to allow you to make your own calls on some of these
+issues. Has your city already completed an LTS analysis? You can plug it
+directly into the BNA, essentially skipping this step entirely. Would you prefer
+to use a more nuanced scheme with five different stress levels? The BNA can
+accommodate that too.
+
+## Block-to-block connectivity analysis
+
+The core of the BNA measures low-stress connections between "blocks". Under the
+default settings, a block is a US Census block, which is roughly analogous to a
+city block. There's no magic to the composition of a block, but coarser areal
+units such as census block groups are less well-suited for use in the BNA since
+their large area can paper over connectivity problems occuring within their
+boundaries.
+
+The process for measuring connectivity can be summarized as follows:
+
+1. Associate roads with blocks (i.e. decide which roads belong to which blocks)
+2. Iterate over each block and identify the shortest route to all other blocks regardless of the stress level
+3. Iterate over each block and identify the shortest route to all other blocks on only low-stress portions of the network
+
+The results of #2 and #3 are then compared. If a connection is found in #3 but
+it requires a significant detour compared to the baseline condition in #2 that
+connection is ignored.
+
+The end result is a matrix of block-to-block connections that identifies which
+blocks are connected to which other blocks via a low-stress route.
+
+## Aggregation of destinations
+
+With the block-to-block connectivity established, the BNA then looks at
+destinations. It starts by measuring the universe of possible destinations
+around each block regardless of stress, and comparing that to the subset of
+destinations accessible via a low-stress route. This process can be summarized:
+
+1. Associate destinations with blocks
+2. Iterate over each block and count the number of destinations accessible regardless of stress level
+3. Iterate over each block and count the number of destinations accessible via a low-stress route
+
+This yields two numbers -- a count of low-stress destinations and a count of all
+destinations -- that can be compared for each category of destination. As the
+number of low-stress destinations approaches the total number of destinations, a
+block's score approaches 100/100. As the number approaches zero, the block's
+score approaches zero too. The score is ultimately dependent on the total number
+of destinations around each blocks. Blocks with very few destinations nearby are
+judged based on how well they connect to those destinations on low-stress
+routes. For example, a block which is connected to two schools will score higher
+if there are only two possible schools to connect to (full points) than if there
+are five possible schools to connect to.
+
+This can be moderated by defining a set thresholds at which points are awarded.
+In the school example above, one could decide that access to two schools is
+sufficient to warrant a 100/100 score regardless of how many additional schools
+are nearby. This is done through an explicit configuration of the school
+destinations in the [configuration file](config.md).
+
+Destination categories are also completely flexible. You may decide that the
+destinations you want to measure access to include coffee shops and movie
+theaters. As long as data can be supplied for a destination type, it can be
+incorporated into BNA results.
+
+The last step is combining the scores for all destination categories into a
+single BNA score. This is accomplished by weights assigned to each destination
+category. A higher weight means that category's score has a larger impact on the
+overall score. One caveat to this can result in surprising scores: if no
+destinations of a given category are identified within reasonable distance of a
+given block, that category's score is not factored into the overall score. For
+example, if a block does not have any universities nearby, the university score
+for that block will be empty and the block's overall score will omit that
+category from consideration in the overall score. This often results in blocks
+in outlying areas receiving very high scores because they are surrounded by few
+destinations, all of which happen to be accessible via a low-stress route. There
+are ways to soften the effects of this with some post-processing of the results
+(e.g. manually reducing scores for blocks who don't meet a certain threshold for
+total number of destinations). As of yet these operations haven't been codified
+into the pyBNA code base.