Col du Clapier 2,482 m · headline 40%

Advocates

Colonel J.-B. Perrin proposed it (Marche d'Annibal des Pyrénées au Pô, 1883; Étude sur Annibal, 1887) after seeing the panorama of the Po plain from the col. Azan (1902, p. 93) and Colin (1904, pp. xiv to xviii and 318 to 321) adopted the col with different approaches to it. Henri Ferrand, the Grenoble alpinist who had doubted the descent, climbed both cols in August 1908 and reported that no point of the Mont-Cenis basin sees the plain (REA 1908, pp. 79 to 84). Wilkinson (1911, pp. 8 to 9) called the view "the crucial evidence". Walbank, looking back on his own work, "hesitatingly came down in favour of the Col du Clapier" (JRS 1956; Commentary 1957; Walbank 2002), as did Huss (1985). Hoyte took an elephant to it in 1959 and crossed by the Mont-Cenis when the Clapier path proved impassable for the animal (1960). Proctor (1971, pp. 210 to 216), Lazenby (1978, pp. 45 to 46), Lancel (1995) and Hunt (2017) followed, Lancel and Hunt preferring the twin saddle of Savine-Coche that de Lavis-Trafford (1956) and Meyer (1964) had proposed. Kuhle and Kuhle defended it against Mahaney (2012, 2015). Hornblower (2024, p. 86): "the best candidate for the pass is the Col de Clapier", because its Italian end is nearest Turin. Bradford (1981) put the odds "slightly in favour" of it over the Traversette.

The case for

A level promontory at the Italian end "which would hold several thousand men, all of whom would at the same time see the view" (Wilkinson, p. 9); what one sees from it, says Colin (p. xv), is "toute l'Italie", not a scrap of plain. Polybius's distances fit an Isère and Arc route "à condition de passer, non pas par le mont Cenis, mais par le Clapier" (Colin, pp. 318 to 321). The Italian end is the nearest of any candidate to the Taurini (Hornblower). The col is high enough for old snow to lie (Lazenby).

The case against

"A lofty and difficult by-pass near the Mont Cenis" (Freshfield 1914, preface). Connolly (1981, p. 165) found no view from the col itself ("only by trekking along the hillside to Monte Aria is this possible"), could not climb the path from the Ambin valley in June 1968 before the snow had cleared, and asked why an army would take it when the lower and easier Mont-Cenis lies beside it. Torr (1924, para 69) saw no motive for Hannibal to go so far north. Bocquet cites the Ambin glacier and the autumn fog of the Po plain. Mahaney finds no rockfall on its descent and no firn.

Here

The viewshed gives the col a 2-degree sliver of plain, 5 per cent of the horizon from 1 km down the descent and 8 per cent from the ridge, so Connolly is right that the saddle itself has no view and Wilkinson is right that the ground beside it has one. The descent by the Clarea is steep and broken. Old snow at the end of October is possible at this height but not usual. The Island at the Isère and the Allobroges at the mouth of the Maurienne fit Polybius without strain; Livy's Druentia does not fit at all. The Clapier loses ground on practicability (Hoyte's elephant, Connolly's June ascent) and on the 1,200 stades, which it undershoots. It leads the headline run at 40 per cent and comes first in 65 draws out of 100 when every range is sampled.

Col de la Traversette 2,947 m · headline 29%

Advocates

Torr (1924, paras 68 to 71) took the army across the Rhône at Tarascon, read Polybius's Skoras as the Durance, placed the Island at the Rhône and Durance confluence and led the army up the Guil to the Traversette. De Beer (Alps and Elephants, 1955; Hannibal's March, 1967; Hannibal, 1969) read Skaras as the Aygues, whose medieval forms are Equeris and Icaris, put the Island at Orange, and took the army by the Drôme, the Col de Grimone, the Durance and the Guil; he thought the Larche "probably the pass which Hannibal intended to cross". Pyatt (1984) and Bagnall (1990) followed him. Prevas (1998; 2017) has crossed the col "well over twenty" times and argues that a pass so remote was unknown to the Romans. Mahaney and colleagues (2007 to 2026) added the rockfall, the firn and the mire. Berti and Vollrath (PNAS, 2026) modelled the energy cost of the crossing and found that "the Col de la Traversette would have been the shortest and energetically most efficient route". Seibert (1993) favoured it, according to Mahaney, and suggested that the army crossed in two columns (Walbank 2002).

The case for

It is the only candidate from which the plain of the Po is seen, and Connolly, who rejects the pass, concedes this (1981, p. 164). Snow lies on it all year in most years (Prevas). Varro's order, with the Montgenèvre as Pompey's pass, puts Hannibal's pass south of the Montgenèvre (Freshfield, Torr, de Beer). Livy's Druentia leads to the Durance basin. The "previous landslide" of Polybius 3.54.7 is matched to the two-tier rockfall below the col, and the churned, Clostridia-rich bed of the mire is dated to the third or second century BC (Mahaney).

The case against

The approach runs through the Combe du Queyras, fifteen kilometres of gorge: "Hannibal would have had to have been a maniac to take his army this way" (Connolly, p. 164), and Azan (pp. 62 to 63) quotes Roman's "l'horreur de cette gorge". The col is more than a thousand metres above the Montgenèvre, and the lower Col de la Croix is passed on the way to it (Connolly). A Royal Engineers survey in 1979 judged it "virtually impossible to get elephants up the Traversette because of the scree or down the other side" (Connolly, p. 165). "Rather than a pass, a narrow breach" (Lancel). Walbank raised philological objections in 1956 (pp. 43 to 45, cited by Hunt; not read here). Kuhle and Kuhle argue from the words Polybius uses for the broken path. There are no Allobroges anywhere near the Guil.

Here

The viewshed confirms the view: 17 per cent of the horizon from the col and 22 to 26 per cent from the knolls beside it. For old snow it ranks highest of any candidate, and its descent is steep and broken. The physical case does not separate the Traversette from the others: the mire's radiocarbon window admits any army from about 380 to 150 BC, and nothing in the published age model's own output places the bed in 218 BC. The pass needs the Island at Orange and an Allobrogian ambush a hundred kilometres from Allobrogian land, and it is the least practicable of the candidates. It is second at 29 per cent, comes first in 29 draws out of 100, and leads at 54 per cent if de Beer's Island is granted.

Mont-Cenis 2,083 m (Petit Mont-Cenis 2,182 m) · headline 16%

Advocates

Simler (1574) named it with the Montgenèvre. Napoleon chose it (Montholon's Mémoires, 1823). Larauza (1826), Maissiat (1874), Osiander (1900) and Jullian (1908, p. 479: the road "plus courte" that "menait sans détour au carrefour de Turin") argued for the Grand Mont-Cenis; Ellis (1853) for the Petit Mont-Cenis, over which de Lavis-Trafford later traced the Roman road (1962). Berthelot (REA 1935 and 1936) put the elephants on it. I found no advocate after Berthelot apart from the general histories that keep it on the shortlist, and Hoyte's party crossed it in 1959 only because the Clapier had failed.

The case for

The easiest col out of the Maurienne, the shortest road to Turin, a plateau with a lake for the summit camp, and Napoleon's choice.

The case against

No point of the basin sees the plain, as Ferrand reported after climbing it in 1908 and as Azan (pp. 58 to 61) and Wilkinson (p. 8, n. 1) had found. The descent is a road, not a cliff path. Old snow does not lie on a 2,083 m col in a normal October.

Here

The viewshed finds no plain from either col or from ten kilometres of either descent. Snow and the broken descent score low; the Island, the Allobroges and the Taurini score as well as they do for the Clapier. It takes 16 per cent, rises to 24 per cent if the view is a rhetorical commonplace, and would lead if the view, the snow and the broken descent were all rhetoric.

Montgenèvre 1,854 m · headline 1.4%

Advocates

Gibbon (1763), Hennebert (Histoire d'Annibal, 1870 to 1891), Bullock Hall (1898), Montanari (REA 1909, "Droit vers le mont Genèvre") and Caspari in the eleventh Britannica argued for it. Connolly (1981, pp. 164 to 166) kept the Island at the Isère and took the army from Grenoble over the Col Bayard or the Col de Manse to the Durance: "the pass was almost certainly the Montgenèvre". Hoyos (Klio 2006; notes to the Oxford Livy, 2006; Mastering the West, 2015, pp. 105 to 106) follows Livy's Tricastini, Vocontii and Tricorii to the Durance and finds that the route "would best lead to the Montgenèvre pass or one of its neighbors such as the Larche". W. T. Arnold, in a letter Freshfield printed (1914, p. 12, n.): "the evidence converges on a Durance pass. The best Durance pass is certainly the Mont Genèvre."

The case for

The lowest and easiest crossing of the western Alps, the road of Hercules, the only road to the Taurini that Polybius is likely to have known (Connolly), the road Livy's tribal itinerary and Ammianus's Durance both lead to, and the road Caesar took in 58 BC.

The case against

Pompey's "iter aliud atque Hannibal" and Varro's separate entries for the two passes count against it (Freshfield 1914, pp. 11 to 12; de Beer; Torr, para 71). There is no view: "Du col du mont Genèvre on n'aperçoit que des sommets de montagnes, pas une plaine, pas même une vallée" (Macé, quoted by Azan, p. 63). It is too low for old snow and too easy for Polybius's descent. Torr (para 68) counts nine days from the Durance to a summit ten miles from it. The Herculean way was exactly the road Scipio expected Hannibal to take (Lancel).

Here

The viewshed finds no plain; for old snow and for a broken descent it ranks lowest of any candidate; the testimonia count against it. Four lines of evidence hold its share down to 1.4 per cent: the view, the old snow, the broken descent, and Varro with Pompey. A Montgenèvre advocate has to read the first three as rhetoric and the fourth as a new road-line in the same corridor, and Connolly and Hoyos do exactly that. If each source family is collapsed to one item it holds 13 per cent.

Col de Larche (Argentière) 1,991 m · headline 3%

Advocates

Freshfield (articles from 1886; Hannibal Once More, 1914, pp. 31 to 57) took the army from the Isère up the Drac to the Durance and the Ubaye, "the Vars-Argentière route", with villages and cornfields all the way. Busquet (1949) and Bailet (1960) argued for it by Barcelonnette. Hoyos (2015) keeps it as the Montgenèvre's neighbour, and de Beer thought it the pass Hannibal meant to cross before the Guil.

The case for

Varro's second pass if his five are read in order from the sea (Freshfield, pp. 20 to 21); an easy, populated route; Livy's Druentia.

The case against

There is no view (Torr, para 69), and the plain is forty miles from the col (Torr, para 55). Wilkinson objected (pp. 45 to 46), and Freshfield's own correspondent preferred the Montgenèvre.

Here

The col has no view of the plain and no old snow, its descent is easy, and its Italian end is far from the Taurini. It scores well on Livy, on Varro and on practicability, which is why it leads at 18 per cent when each source family is collapsed to one item. Headline share: 3 per cent.

Petit St Bernard 2,188 m · headline 1.5%

Advocates

De Luc (1818), Wickham and Cramer (1820), Law (1866), Thomas Arnold, Mommsen, Dodge (1891), Lenschau in the Realencyclopädie, Hyde (1935), Pais, de Coninck (1992), Debelmas and Bocquet, who adds an Allobrogian coin type of Punic design and rejects the view criterion because of the Po fog.

The case for

Coelius's Cremonis iugum (Livy 21.38.7), Nepos's saltus Graius, the Ceutrones' wagon road (Strabo 4.6.11), the wealth of Allobrogian land for supply, and the Isère followed to its head.

The case against

Livy's own objection: the pass leads down to the Libui and the Salassi, not the Taurini (21.38.7 to 8). Polybius's list of four passes (34.10.18) names the Taurini one for Hannibal. There is no view, and the plain is seventy-five miles from the col (Torr, para 55; Colin, p. 321: "d'affreux précipices").

Here

The Island and the Allobroges fit as well as for any Isère col. The Taurini, the view, the snow and the descent all count against it. Headline share: 1.5 per cent.

Grand St Bernard 2,469 m · headline 0.3%

Advocates

The "vulgar" Roman tradition that named the Pennine Alps from the Poeni (Livy 21.38.6; Pliny 3.123; Servius), and Whitaker (1794).

The case against

Livy's refutation, Servius's etymology from the god Poeninus, and Polybius's list; Torr (para 23) found it "incredible that Hannibal should have crossed a pass so far eastward". It stays in the table so that the Roman tradition is weighed.

Another col headline 9%

Candidates

The Col de Malaure, whose view is "from a narrow ridge on which no body of troops could be stationed" (Wilkinson, p. 9) and which Cottrell (1961) shortlisted with the Traversette and the Clapier; the Agnel and the Col de la Croix on the Guil side; the Autaret and the other Maurienne cols; the Belledonne short cut of de Galbert (2005), which leads to the Maurienne cols rather than to a new one; and Seibert's two columns.

Here

This row stands for the cols nobody has argued for at length. It carries 9 per cent because Polybius's description would fit an unnamed col of the Cottian range about as well as it fits the named ones, and a model that gave such a col nothing would claim more for the named candidates than they have established.