Three numbers from FHWA’s own tables, taken in sequence, contradict most of what gets said about American bridges. In 2018 the National Bridge Inventory held 616,096 bridges, of which 47,054 were classified Poor, a share of 7.6 percent. In 2021 it held 619,622, of which 43,586 were Poor, a share of 7.0 percent. In 2025 it held 624,193, of which 41,685 were Poor, a share of 6.7 percent, with Guam not submitting data that year.
The inventory grew by roughly 8,000 bridges over that period and the number in Poor condition fell by more than 5,000. Both the count and the share are moving in the same direction, and the direction is down. The 2025 table also splits the National Highway System out: 147,633 NHS bridges, of which 4,341 are Poor, a share of 2.9 percent, against 63,757 Good and 79,535 Fair. The busiest part of the network is in materially better condition than the network as a whole.
Why the federal measure weights deck area, not bridge count
Every figure above counts bridges. The measures that actually bind state departments of transportation do not. Most of the condition arguments collected in the bridges and structures section sit on top of that split.
Under 23 CFR 490.409, the percentage of bridges in Good condition and the percentage in Poor condition on the National Highway System are computed as the sum of length multiplied by width for the bridges in each class, divided by the same sum for all NHS bridges, reported to a tenth of a percent. A 40-foot rural overpass and a two-mile estuary crossing are one bridge each in the count and are separated by three orders of magnitude in the measure. The classification itself comes from the minimum condition rating method described in how bridges are inspected and rated, and of the three classes only Good and Poor are performance measures under 490.407. Fair is a residual.
The consequences of the two denominators are visible in the same federal document. FHWA’s 25th edition Conditions and Performance report to Congress, which draws primarily on 2018 data and was submitted in February 2024, states that on unweighted bridge count the share of bridges classified as poor dropped from 10.1 percent in 2008 to 7.6 percent in 2018, while the share classified as good rose from 46.0 percent to 47.8 percent. Weighted by deck area, over the same decade, the poor share declined from 8.8 percent to 5.4 percent, and the poor share of NHS bridges specifically dropped from 8.0 percent to 4.5 percent.
Then the good share does something the count does not. Weighted by deck area, the report says, the share classified as good declined slightly, from 45.8 percent in 2008 to 45.3 percent in 2018, even as the NHS good share improved from 43.1 percent to 43.4 percent. Read on count, good conditions improved. Read on deck area, good conditions on the full inventory went backwards while poor conditions improved sharply. Both statements come from the same report, and the reconciliation is that large bridges left the poor category faster than they entered the good one. They moved to fair.
What Poor does not mean
The report states the point in a box, and it is the sentence most often missing from coverage of these figures.
The classification of a bridge as poor does not mean it is unsafe; bridges that are considered unsafe are closed to traffic.
The same report explains why the weighting exists at all: classifications are often weighted by bridge deck area, because in general larger bridges are costlier to rehabilitate or replace than smaller ones. The weighting is a cost proxy, not a safety proxy, and it is the right choice for a measure whose purpose is to allocate money. It is the wrong choice for a headline about how many bridges are in trouble, and the two get mixed constantly.
A related definition changed under the same regulation. Beginning with calendar year 2018, 23 CFR 490.411 defines a structurally deficient bridge as one with any component in Poor or worse condition. Before 2018 the definition also captured bridges whose structural evaluation or waterway adequacy appraisal ratings were 2 or less. Any time series that crosses 2018 without saying so is comparing two different definitions, and the earlier one was broader. The primary tables that carry each of these figures, and the trap sitting inside each one, are catalogued in key FHWA data sources.
Where the backlog number comes from
The figure in circulation is $191 billion, and it is traceable. The 2025 ASCE report card puts system rehabilitation needs for bridges at $191.3 billion, and separately notes that FHWA’s most recent assessment, which ASCE dates to 2018, put $1.1 trillion against an existing backlog of highway and bridge investments.
FHWA’s own 25th edition report shows how the pieces fit. Under its Improve Conditions and Performance scenario, about 36.1 percent of capital investment would go to addressing a backlog of cost-beneficial investments of $1.1 trillion. That total splits into $237 billion for system expansion and $852 billion for existing assets, and the $852 billion Highway Repair Backlog comprises $511 billion for the pavement component of system rehabilitation, $191 billion for the bridge component, and $150 billion for system enhancement. All of it is stated in constant 2018 dollars, over a prospective analysis period running from 2018 to 2038.
Three features of that arithmetic get lost in the citation. The backlog is defined as cost-beneficial investments, meaning the model has already discarded work whose benefits do not exceed its costs, so the number is smaller than a wish list and larger than a maintenance budget. The bridge component is a rehabilitation figure and sits inside a repair backlog that is mostly pavement. And the base year is 2018, which puts the whole estimate on travel and price conditions from before the pandemic.
The assumptions inside the estimate
The bridge half of that estimate is produced by the National Bridge Investment Analysis System, and FHWA documents its method in an appendix that is more candid than the headline number suggests. NBIAS first appeared in the 1999 edition of the report. It does not read a full inventory of structural elements. Instead it uses a set of synthesis, quantity and condition models to predict which elements exist on each bridge in the inventory and what condition they are in, using a decision tree and Monte Carlo simulation, with deterioration rates that vary across nine climate zones.
The appendix then states something that anyone quoting the total should know: the current version of NBIAS can accept the direct import of structural element data where such data are available, but that capability was not used in the development of the report. Element-level data collection is required for National Highway System bridges under the inspection standards, and the needs model runs on synthesised elements anyway.
The same appendix records the model’s other limits without euphemism. Its maintenance, repair and rehabilitation algorithms were last fully recalibrated in 2006 and, in FHWA’s own words, are no longer fully consistent with current bridge management practices. The functional improvement standards the model applies by functional class are derived in part from sufficiency rating calculations, so the metric that the 2022 inventory specification discontinued still shapes the federal needs estimate. When NBIAS selects a structure for replacement it replaces it with one of the same type and capacity, irrespective of whether added capacity is needed. And of roughly 125,000 culverts in the inventory, the model does not contain the algorithms needed to conduct a full analysis, so their needs are handled by approximation.
None of that makes $191 billion wrong. It makes it a model output with a stated vintage and stated gaps, which is a different object from a repair bill. What the choice between repairing and replacing looks like on an individual structure is a separate question, examined in how agencies decide between rehab and replacement.
One discrepancy in the report itself
The 25th edition’s own text and its exhibit disagree on what is being counted. In the passage projecting outcomes under the Improve Conditions and Performance scenario, the narrative says the share of bridges classified as poor is projected to improve, decreasing from 5.4 percent in 2018 to 1.2 percent in 2038. The accompanying exhibit labels that same series Share of Deck Area on Poor Bridges. The 5.4 percent figure is the deck-area-weighted value quoted earlier in the report, so the exhibit label is the accurate one and the sentence uses the looser phrasing. The projection is about deck area.
The report also declines to promise elimination, and explains why in terms of its own decision rule: the scenario would not eliminate all poor pavements and bridges because in some cases improving assets becomes cost-beneficial only after they have declined into poor condition, and in others improving them before they reach poor condition is cost-beneficial. A benefit-cost model will leave some structures in poor condition on purpose.
For anyone citing these figures, the arithmetic that matters is the denominator. Poor is 6.7 percent of bridges and 5.4 percent of deck area, 2.9 percent of NHS bridges by count and 4.5 percent of NHS deck area, the count figures for 2025 and the deck-area figures for 2018. Each of those four numbers is correct and none of them substitutes for another.